var bibbase_data = {"data":"\"Loading..\"\n\n
\n\n \n\n \n\n \n \n\n \n\n \n \n\n \n\n \n
\n generated by\n \n \"bibbase.org\"\n\n \n
\n \n\n
\n\n \n\n\n
\n\n Excellent! Next you can\n create a new website with this list, or\n embed it in an existing web page by copying & pasting\n any of the following snippets.\n\n
\n JavaScript\n (easiest)\n
\n \n <script src=\"https://bibbase.org/show?bib=https://www.shane.st/teaching/575/spr24/575_Compositionality.bib&jsonp=1&theme=dividers&urlLinks=URL&jsonp=1\"></script>\n \n
\n\n PHP\n
\n \n <?php\n $contents = file_get_contents(\"https://bibbase.org/show?bib=https://www.shane.st/teaching/575/spr24/575_Compositionality.bib&jsonp=1&theme=dividers&urlLinks=URL\");\n print_r($contents);\n ?>\n \n
\n\n iFrame\n (not recommended)\n
\n \n <iframe src=\"https://bibbase.org/show?bib=https://www.shane.st/teaching/575/spr24/575_Compositionality.bib&jsonp=1&theme=dividers&urlLinks=URL\"></iframe>\n \n
\n\n

\n For more details see the documention.\n

\n
\n
\n\n
\n\n This is a preview! To use this list on your own web site\n or create a new web site from it,\n create a free account. The file will be added\n and you will be able to edit it in the File Manager.\n We will show you instructions once you've created your account.\n
\n\n
\n\n

To the site owner:

\n\n

Action required! Mendeley is changing its\n API. In order to keep using Mendeley with BibBase past April\n 14th, you need to:\n

    \n
  1. renew the authorization for BibBase on Mendeley, and
    2. \n
  2. update the BibBase URL\n in your page the same way you did when you initially set up\n this page.\n
    3. \n
\n

\n\n

\n \n \n Fix it now\n

\n
\n\n
\n\n\n
\n \n \n
\n
\n  \n 2024\n \n \n (1)\n \n \n
\n
\n \n \n
\n \n\n \n \n \n \n \n \n Minimal Compositionality versus Bird Implicatures: Two Theories of ABC-D Sequences in Japanese Tits.\n \n \n \n \n\n\n \n Schlenker, P.; Salis, A.; Leroux, M.; Coye, C.; Rizzi, L.; Steinert-Threlkeld, S.; and Chemla, E.\n\n\n \n\n\n\n Biological Reviews. 2024.\n \n\n\n\n
\n\n\n\n \n \n \"MinimalPaper\n  \n \n\n \n \n doi\n  \n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n  \n \n 2 downloads\n \n \n\n \n \n \n \n \n \n \n\n  \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n\n\n\n
\n 
@article{schlenkerMinimalCompositionalityBird2024,\n  title = {Minimal {{Compositionality}} versus {{Bird Implicatures}}:  {{Two Theories}} of {{ABC-D Sequences}} in {{Japanese Tits}}},\n  shorttitle = {Minimal {{Compositionality}} versus {{Bird Implicatures}}},\n  author = {Schlenker, Philippe and Salis, Ambre and Leroux, Ma{\\"e}l and Coye, Camille and Rizzi, Luigi and {Steinert-Threlkeld}, Shane and Chemla, Emmanuel},\n  year = {2024},\n  journal = {Biological Reviews},\n  doi = {10.1111/brv.13068},\n  url = {https://ling.auf.net/lingbuzz/007422},\n  urldate = {2024-02-26},\n  abstract = {It was argued in a series of experimental studies that Japanese tits (Parus minor) have an ABC call that has an alert function, a D call that has a recruitment function, and an ABC-D call that is compositionally derived from ABC and D, and has a mobbing function. A key conclusion was that ABC-D differs from the combination of separate utterances of ABC and of D (e.g. as played by distinct but close loudspeakers). While the logic of the argument is arguably sound, no explicit rule has been proposed to derive the meaning of ABC-D from that of its parts. We compare two analyses. One posits a limited instance of semantic compositionality ('Minimal Compositionality'); the other does without compositionality, but with a more sophisticated pragmatics ('Bird Implicatures'). Minimal Compositionality takes the composition of ABC and D to deviate only minimally from what would be found with two independent utterances: ABC means that 'there is something that licenses an alert', D means that 'there is something that licenses recruitment', and ABC-D means that 'there is something that licenses both an alert and recruitment'. By contrast, ABC and D as independent utterances yield something weaker, namely: 'there is something that licenses an alert, and there is something that licenses recruitment', without any 'binding' across the two utterances. The second theory, Bird Implicatures, only requires that ABC-D should be more informative than ABC, and/or than D. It builds on the idea, proposed for several monkey species, that a less informative call competes with a more informative one (`Informativity Principle'): when produced alone, ABC and D trigger an inference that ABC-D is false. We explain how both Minimal Compositionality and Bird Implicatures could have evolved, and we compare the predictions of the two theories. Finally, we extend the discussion to some chimpanzee and meerkat sequences that might raise related theoretical problems.},\n  keywords = {animal linguistics,animal semantics,bird calls,compositionality,implicatures,informativity principle,meerkat calls,minimal compositionality,morphology,semantics,syntax},\n  annotation = {LingBuzz Published In: To appear in Biological Reviews},\n  file = {/Users/shanest/sync/library/Schlenker et al/2024/Schlenker et al. - 2024 - Minimal Compositionality versus Bird Implicatures.pdf;/Users/shanest/Zotero/storage/DW8DDL8T/007422.html}\n}\n\n
\n
\n\n\n
\n It was argued in a series of experimental studies that Japanese tits (Parus minor) have an ABC call that has an alert function, a D call that has a recruitment function, and an ABC-D call that is compositionally derived from ABC and D, and has a mobbing function. A key conclusion was that ABC-D differs from the combination of separate utterances of ABC and of D (e.g. as played by distinct but close loudspeakers). While the logic of the argument is arguably sound, no explicit rule has been proposed to derive the meaning of ABC-D from that of its parts. We compare two analyses. One posits a limited instance of semantic compositionality ('Minimal Compositionality'); the other does without compositionality, but with a more sophisticated pragmatics ('Bird Implicatures'). Minimal Compositionality takes the composition of ABC and D to deviate only minimally from what would be found with two independent utterances: ABC means that 'there is something that licenses an alert', D means that 'there is something that licenses recruitment', and ABC-D means that 'there is something that licenses both an alert and recruitment'. By contrast, ABC and D as independent utterances yield something weaker, namely: 'there is something that licenses an alert, and there is something that licenses recruitment', without any 'binding' across the two utterances. The second theory, Bird Implicatures, only requires that ABC-D should be more informative than ABC, and/or than D. It builds on the idea, proposed for several monkey species, that a less informative call competes with a more informative one (`Informativity Principle'): when produced alone, ABC and D trigger an inference that ABC-D is false. We explain how both Minimal Compositionality and Bird Implicatures could have evolved, and we compare the predictions of the two theories. Finally, we extend the discussion to some chimpanzee and meerkat sequences that might raise related theoretical problems.\n
\n\n\n
\n\n\n\n\n\n
\n
\n\n
\n
\n  \n 2023\n \n \n (12)\n \n \n
\n
\n \n \n
\n \n\n \n \n \n \n \n \n How Do In-Context Examples Affect Compositional Generalization?.\n \n \n \n \n\n\n \n An, S.; Lin, Z.; Fu, Q.; Chen, B.; Zheng, N.; Lou, J.; and Zhang, D.\n\n\n \n\n\n\n In Rogers, A.; Boyd-Graber, J.; and Okazaki, N., editor(s), Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 11027–11052, Toronto, Canada, July 2023. Association for Computational Linguistics\n \n\n\n\n
\n\n\n\n \n \n \"HowPaper\n  \n \n\n \n \n doi\n  \n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n  \n \n 1 download\n \n \n\n \n \n \n \n \n \n \n\n  \n \n \n\n\n\n
\n 
@inproceedings{anHowInContextExamples2023,\n  title = {How {{Do In-Context Examples Affect Compositional Generalization}}?},\n  booktitle = {Proceedings of the 61st {{Annual Meeting}} of the {{Association}} for {{Computational Linguistics}} ({{Volume}} 1: {{Long Papers}})},\n  author = {An, Shengnan and Lin, Zeqi and Fu, Qiang and Chen, Bei and Zheng, Nanning and Lou, Jian-Guang and Zhang, Dongmei},\n  editor = {Rogers, Anna and {Boyd-Graber}, Jordan and Okazaki, Naoaki},\n  year = {2023},\n  month = jul,\n  pages = {11027--11052},\n  publisher = {Association for Computational Linguistics},\n  address = {Toronto, Canada},\n  doi = {10.18653/v1/2023.acl-long.618},\n  url = {https://aclanthology.org/2023.acl-long.618},\n  urldate = {2024-03-19},\n  abstract = {Compositional generalization--understanding unseen combinations of seen primitives--is an essential reasoning capability in human intelligence. The AI community mainly studies this capability by fine-tuning neural networks on lots of training samples, while it is still unclear whether and how in-context learning--the prevailing few-shot paradigm based on large language models--exhibits compositional generalization. In this paper, we present CoFe, a test suite to investigate in-context compositional generalization. We find that the compositional generalization performance can be easily affected by the selection of in-context examples, thus raising the research question what the key factors are to make good in-context examples for compositional generalization. We study three potential factors: similarity, diversity and complexity. Our systematic experiments indicate that in-context examples should be structurally similar to the test case, diverse from each other, and individually simple. Furthermore, two strong limitations are observed: in-context compositional generalization on fictional words is much weaker than that on commonly used ones; it is still critical that the in-context examples should cover required linguistic structures, even though the backbone model has been pre-trained on large corpus. We hope our analysis would facilitate the understanding and utilization of in-context learning paradigm.},\n  file = {/Users/shanest/sync/library/An et al/2023/An et al. - 2023 - How Do In-Context Examples Affect Compositional Ge.pdf}\n}\n\n
\n
\n\n\n
\n Compositional generalization–understanding unseen combinations of seen primitives–is an essential reasoning capability in human intelligence. The AI community mainly studies this capability by fine-tuning neural networks on lots of training samples, while it is still unclear whether and how in-context learning–the prevailing few-shot paradigm based on large language models–exhibits compositional generalization. In this paper, we present CoFe, a test suite to investigate in-context compositional generalization. We find that the compositional generalization performance can be easily affected by the selection of in-context examples, thus raising the research question what the key factors are to make good in-context examples for compositional generalization. We study three potential factors: similarity, diversity and complexity. Our systematic experiments indicate that in-context examples should be structurally similar to the test case, diverse from each other, and individually simple. Furthermore, two strong limitations are observed: in-context compositional generalization on fictional words is much weaker than that on commonly used ones; it is still critical that the in-context examples should cover required linguistic structures, even though the backbone model has been pre-trained on large corpus. We hope our analysis would facilitate the understanding and utilization of in-context learning paradigm.\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n \n Compositional Semantic Parsing with Large Language Models.\n \n \n \n \n\n\n \n Drozdov, A.; Schärli, N.; Akyürek, E.; Scales, N.; Song, X.; Chen, X.; Bousquet, O.; and Zhou, D.\n\n\n \n\n\n\n In The Eleventh International Conference on Learning Representations, 2023. \n \n\n\n\n
\n\n\n\n \n \n \"CompositionalPaper\n  \n \n\n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n  \n \n 11 downloads\n \n \n\n \n \n \n \n \n \n \n\n  \n \n \n\n\n\n
\n 
@inproceedings{drozdovCompositionalSemanticParsing2023,\n  title = {Compositional {{Semantic Parsing}} with {{Large Language Models}}},\n  booktitle = {The {{Eleventh International Conference}} on {{Learning Representations}}},\n  author = {Drozdov, Andrew and Sch{\\"a}rli, Nathanael and Aky{\\"u}rek, Ekin and Scales, Nathan and Song, Xinying and Chen, Xinyun and Bousquet, Olivier and Zhou, Denny},\n  year = {2023},\n  url = {https://openreview.net/forum?id=gJW8hSGBys8},\n  urldate = {2024-03-19},\n  abstract = {Humans can reason compositionally when presented with new tasks. Previous research shows that appropriate prompting techniques enable large language models (LLMs) to solve artificial compositional generalization tasks such as SCAN. In this work, we identify additional challenges in more realistic semantic parsing tasks with larger vocabulary and refine these prompting techniques to address them. Our best method is based on least-to-most prompting: it decomposes the problem using prompting-based syntactic parsing, then uses this decomposition to select appropriate exemplars and to sequentially generate the semantic parse. This method allows us to set a new state of the art for CFQ while requiring only 1\\% of the training data used by traditional approaches. Due to the general nature of our approach, we expect similar efforts will lead to new results in other tasks and domains, especially for knowledge-intensive applications.},\n  langid = {english},\n  file = {/Users/shanest/sync/library/Drozdov et al/2022/Drozdov et al. - 2022 - Compositional Semantic Parsing with Large Language.pdf}\n}\n\n
\n
\n\n\n
\n Humans can reason compositionally when presented with new tasks. Previous research shows that appropriate prompting techniques enable large language models (LLMs) to solve artificial compositional generalization tasks such as SCAN. In this work, we identify additional challenges in more realistic semantic parsing tasks with larger vocabulary and refine these prompting techniques to address them. Our best method is based on least-to-most prompting: it decomposes the problem using prompting-based syntactic parsing, then uses this decomposition to select appropriate exemplars and to sequentially generate the semantic parse. This method allows us to set a new state of the art for CFQ while requiring only 1% of the training data used by traditional approaches. Due to the general nature of our approach, we expect similar efforts will lead to new results in other tasks and domains, especially for knowledge-intensive applications.\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n \n Faith and Fate: Limits of Transformers on Compositionality.\n \n \n \n \n\n\n \n Dziri, N.; Lu, X.; Sclar, M.; Li, X. L.; Jiang, L.; Lin, B. Y.; Welleck, S.; West, P.; Bhagavatula, C.; Bras, R. L.; Hwang, J. D.; Sanyal, S.; Ren, X.; Ettinger, A.; Harchaoui, Z.; and Choi, Y.\n\n\n \n\n\n\n In Thirty-Seventh Conference on Neural Information Processing Systems, November 2023. \n \n\n\n\n
\n\n\n\n \n \n \"FaithPaper\n  \n \n\n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n  \n \n 8 downloads\n \n \n\n \n \n \n \n \n \n \n\n  \n \n \n\n\n\n
\n 
@inproceedings{dziriFaithFateLimits2023,\n  title = {Faith and {{Fate}}: {{Limits}} of {{Transformers}} on {{Compositionality}}},\n  shorttitle = {Faith and {{Fate}}},\n  booktitle = {Thirty-Seventh {{Conference}} on {{Neural Information Processing Systems}}},\n  author = {Dziri, Nouha and Lu, Ximing and Sclar, Melanie and Li, Xiang Lorraine and Jiang, Liwei and Lin, Bill Yuchen and Welleck, Sean and West, Peter and Bhagavatula, Chandra and Bras, Ronan Le and Hwang, Jena D. and Sanyal, Soumya and Ren, Xiang and Ettinger, Allyson and Harchaoui, Zaid and Choi, Yejin},\n  year = {2023},\n  month = nov,\n  url = {https://openreview.net/forum?id=Fkckkr3ya},\n  urldate = {2024-03-18},\n  abstract = {Transformer large language models (LLMs) have sparked admiration for their exceptional performance on tasks that demand intricate multi-step reasoning. Yet, these models simultaneously show failures on surprisingly trivial problems. This begs the question: Are these errors incidental, or do they signal more substantial limitations? In an attempt to demystify transformer LLMs, we investigate the limits of these models across three representative compositional tasks---multi-digit multiplication, logic grid puzzles, and a classic dynamic programming problem. These tasks require breaking problems down into sub-steps and synthesizing these steps into a precise answer. We formulate compositional tasks as computation graphs to systematically quantify the level of complexity, and break down reasoning steps into intermediate sub-procedures. Our empirical findings suggest that transformer LLMs solve compositional tasks by reducing multi-step compositional reasoning into linearized subgraph matching, without necessarily developing systematic problem-solving skills. To round off our empirical study, we provide theoretical arguments on abstract multi-step reasoning problems that highlight how autoregressive generations' performance can rapidly decay with increased task complexity.},\n  langid = {english},\n  file = {/Users/shanest/sync/library/Dziri et al/2023/Dziri et al. - 2023 - Faith and Fate Limits of Transformers on Composit.pdf}\n}\n\n
\n
\n\n\n
\n Transformer large language models (LLMs) have sparked admiration for their exceptional performance on tasks that demand intricate multi-step reasoning. Yet, these models simultaneously show failures on surprisingly trivial problems. This begs the question: Are these errors incidental, or do they signal more substantial limitations? In an attempt to demystify transformer LLMs, we investigate the limits of these models across three representative compositional tasks—multi-digit multiplication, logic grid puzzles, and a classic dynamic programming problem. These tasks require breaking problems down into sub-steps and synthesizing these steps into a precise answer. We formulate compositional tasks as computation graphs to systematically quantify the level of complexity, and break down reasoning steps into intermediate sub-procedures. Our empirical findings suggest that transformer LLMs solve compositional tasks by reducing multi-step compositional reasoning into linearized subgraph matching, without necessarily developing systematic problem-solving skills. To round off our empirical study, we provide theoretical arguments on abstract multi-step reasoning problems that highlight how autoregressive generations' performance can rapidly decay with increased task complexity.\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n \n A Taxonomy and Review of Generalization Research in NLP.\n \n \n \n \n\n\n \n Hupkes, D.; Giulianelli, M.; Dankers, V.; Artetxe, M.; Elazar, Y.; Pimentel, T.; Christodoulopoulos, C.; Lasri, K.; Saphra, N.; Sinclair, A.; Ulmer, D.; Schottmann, F.; Batsuren, K.; Sun, K.; Sinha, K.; Khalatbari, L.; Ryskina, M.; Frieske, R.; Cotterell, R.; and Jin, Z.\n\n\n \n\n\n\n Nature Machine Intelligence, 5(10): 1161–1174. October 2023.\n \n\n\n\n
\n\n\n\n \n \n \"APaper\n  \n \n\n \n \n doi\n  \n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n  \n \n 3 downloads\n \n \n\n \n \n \n \n \n \n \n\n  \n \n \n \n \n \n \n\n\n\n
\n 
@article{hupkesTaxonomyReviewGeneralization2023,\n  title = {A Taxonomy and Review of Generalization Research in {{NLP}}},\n  author = {Hupkes, Dieuwke and Giulianelli, Mario and Dankers, Verna and Artetxe, Mikel and Elazar, Yanai and Pimentel, Tiago and Christodoulopoulos, Christos and Lasri, Karim and Saphra, Naomi and Sinclair, Arabella and Ulmer, Dennis and Schottmann, Florian and Batsuren, Khuyagbaatar and Sun, Kaiser and Sinha, Koustuv and Khalatbari, Leila and Ryskina, Maria and Frieske, Rita and Cotterell, Ryan and Jin, Zhijing},\n  year = {2023},\n  month = oct,\n  journal = {Nature Machine Intelligence},\n  volume = {5},\n  number = {10},\n  pages = {1161--1174},\n  publisher = {Nature Publishing Group},\n  issn = {2522-5839},\n  doi = {10.1038/s42256-023-00729-y},\n  url = {https://www.nature.com/articles/s42256-023-00729-y},\n  urldate = {2024-03-18},\n  abstract = {The ability to generalize well is one of the primary desiderata for models of natural language processing (NLP), but what `good generalization' entails and how it should be evaluated is not well understood. In this Analysis we present a taxonomy for characterizing and understanding generalization research in NLP. The proposed taxonomy is based on an extensive literature review and contains five axes along which generalization studies can differ: their main motivation, the type of generalization they aim to solve, the type of data shift they consider, the source by which this data shift originated, and the locus of the shift within the NLP modelling pipeline. We use our taxonomy to classify over 700 experiments, and we use the results to present an in-depth analysis that maps out the current state of generalization research in NLP and make recommendations for which areas deserve attention in the future.},\n  copyright = {2023 The Author(s)},\n  langid = {english},\n  keywords = {Computer science,Language and linguistics},\n  file = {/Users/shanest/sync/library/Hupkes et al/2023/Hupkes et al. - 2023 - A taxonomy and review of generalization research i.pdf}\n}\n\n
\n
\n\n\n
\n The ability to generalize well is one of the primary desiderata for models of natural language processing (NLP), but what `good generalization' entails and how it should be evaluated is not well understood. In this Analysis we present a taxonomy for characterizing and understanding generalization research in NLP. The proposed taxonomy is based on an extensive literature review and contains five axes along which generalization studies can differ: their main motivation, the type of generalization they aim to solve, the type of data shift they consider, the source by which this data shift originated, and the locus of the shift within the NLP modelling pipeline. We use our taxonomy to classify over 700 experiments, and we use the results to present an in-depth analysis that maps out the current state of generalization research in NLP and make recommendations for which areas deserve attention in the future.\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n \n Human-like Systematic Generalization through a Meta-Learning Neural Network.\n \n \n \n \n\n\n \n Lake, B. M.; and Baroni, M.\n\n\n \n\n\n\n Nature, 623(7985): 115–121. November 2023.\n \n\n\n\n
\n\n\n\n \n \n \"Human-likePaper\n  \n \n\n \n \n doi\n  \n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n  \n \n 10 downloads\n \n \n\n \n \n \n \n \n \n \n\n  \n \n \n \n \n \n \n\n\n\n
\n 
@article{lakeHumanlikeSystematicGeneralization2023a,\n  title = {Human-like Systematic Generalization through a Meta-Learning Neural Network},\n  author = {Lake, Brenden M. and Baroni, Marco},\n  year = {2023},\n  month = nov,\n  journal = {Nature},\n  volume = {623},\n  number = {7985},\n  pages = {115--121},\n  publisher = {Nature Publishing Group},\n  issn = {1476-4687},\n  doi = {10.1038/s41586-023-06668-3},\n  url = {https://www.nature.com/articles/s41586-023-06668-3},\n  urldate = {2024-03-18},\n  abstract = {The power of human language and thought arises from systematic compositionality---the algebraic ability to understand and produce novel combinations from known components. Fodor and Pylyshyn1 famously argued that artificial neural networks lack this capacity and are therefore not viable models of the mind. Neural networks have advanced considerably in the years since, yet the systematicity challenge persists. Here we successfully address Fodor and Pylyshyn's challenge by providing evidence that neural networks can achieve human-like systematicity when optimized for their compositional skills. To do so, we introduce the meta-learning for compositionality (MLC) approach for guiding training through a dynamic stream of compositional tasks. To compare humans and machines, we conducted human behavioural experiments using an~instruction learning paradigm. After considering seven different models, we found that, in contrast to perfectly systematic but rigid probabilistic symbolic models, and perfectly flexible but unsystematic neural networks, only MLC achieves both the systematicity and flexibility needed for human-like generalization. MLC also advances the compositional skills of machine learning systems in several systematic generalization benchmarks. Our results show how a standard neural network architecture, optimized for its compositional skills, can mimic human systematic generalization in a head-to-head comparison.},\n  copyright = {2023 The Author(s)},\n  langid = {english},\n  keywords = {Computer science,Human behaviour},\n  file = {/Users/shanest/sync/library/Lake_Baroni/2023/Lake and Baroni - 2023 - Human-like systematic generalization through a met.pdf}\n}\n\n
\n
\n\n\n
\n The power of human language and thought arises from systematic compositionality—the algebraic ability to understand and produce novel combinations from known components. Fodor and Pylyshyn1 famously argued that artificial neural networks lack this capacity and are therefore not viable models of the mind. Neural networks have advanced considerably in the years since, yet the systematicity challenge persists. Here we successfully address Fodor and Pylyshyn's challenge by providing evidence that neural networks can achieve human-like systematicity when optimized for their compositional skills. To do so, we introduce the meta-learning for compositionality (MLC) approach for guiding training through a dynamic stream of compositional tasks. To compare humans and machines, we conducted human behavioural experiments using an instruction learning paradigm. After considering seven different models, we found that, in contrast to perfectly systematic but rigid probabilistic symbolic models, and perfectly flexible but unsystematic neural networks, only MLC achieves both the systematicity and flexibility needed for human-like generalization. MLC also advances the compositional skills of machine learning systems in several systematic generalization benchmarks. Our results show how a standard neural network architecture, optimized for its compositional skills, can mimic human systematic generalization in a head-to-head comparison.\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n \n Compositional Generalisation with Structured Reordering and Fertility Layers.\n \n \n \n \n\n\n \n Lindemann, M.; Koller, A.; and Titov, I.\n\n\n \n\n\n\n In Vlachos, A.; and Augenstein, I., editor(s), Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics, pages 2172–2186, Dubrovnik, Croatia, May 2023. Association for Computational Linguistics\n \n\n\n\n
\n\n\n\n \n \n \"CompositionalPaper\n  \n \n\n \n \n doi\n  \n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n\n\n\n
\n 
@inproceedings{lindemannCompositionalGeneralisationStructured2023,\n  title = {Compositional {{Generalisation}} with {{Structured Reordering}} and {{Fertility Layers}}},\n  booktitle = {Proceedings of the 17th {{Conference}} of the {{European Chapter}} of the {{Association}} for {{Computational Linguistics}}},\n  author = {Lindemann, Matthias and Koller, Alexander and Titov, Ivan},\n  editor = {Vlachos, Andreas and Augenstein, Isabelle},\n  year = {2023},\n  month = may,\n  pages = {2172--2186},\n  publisher = {Association for Computational Linguistics},\n  address = {Dubrovnik, Croatia},\n  doi = {10.18653/v1/2023.eacl-main.159},\n  url = {https://aclanthology.org/2023.eacl-main.159},\n  urldate = {2024-04-02},\n  abstract = {Seq2seq models have been shown to struggle with compositional generalisation, i.e. generalising to new and potentially more complex structures than seen during training. Taking inspiration from grammar-based models that excel at compositional generalisation, we present a flexible end-to-end differentiable neural model that composes two structural operations: a fertility step, which we introduce in this work, and a reordering step based on previous work (Wang et al., 2021). To ensure differentiability, we use the expected value of each step, which we compute using dynamic programming. Our model outperforms seq2seq models by a wide margin on challenging compositional splits of realistic semantic parsing tasks that require generalisation to longer examples. It also compares favourably to other models targeting compositional generalisation.}\n}\n\n
\n
\n\n\n
\n Seq2seq models have been shown to struggle with compositional generalisation, i.e. generalising to new and potentially more complex structures than seen during training. Taking inspiration from grammar-based models that excel at compositional generalisation, we present a flexible end-to-end differentiable neural model that composes two structural operations: a fertility step, which we introduce in this work, and a reordering step based on previous work (Wang et al., 2021). To ensure differentiability, we use the expected value of each step, which we compute using dynamic programming. Our model outperforms seq2seq models by a wide margin on challenging compositional splits of realistic semantic parsing tasks that require generalisation to longer examples. It also compares favourably to other models targeting compositional generalisation.\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n \n Compositional Generalization without Trees Using Multiset Tagging and Latent Permutations.\n \n \n \n \n\n\n \n Lindemann, M.; Koller, A.; and Titov, I.\n\n\n \n\n\n\n In Rogers, A.; Boyd-Graber, J.; and Okazaki, N., editor(s), Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 14488–14506, Toronto, Canada, July 2023. Association for Computational Linguistics\n \n\n\n\n
\n\n\n\n \n \n \"CompositionalPaper\n  \n \n\n \n \n doi\n  \n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n\n\n\n
\n 
@inproceedings{lindemannCompositionalGeneralizationTrees2023,\n  title = {Compositional {{Generalization}} without {{Trees}} Using {{Multiset Tagging}} and {{Latent Permutations}}},\n  booktitle = {Proceedings of the 61st {{Annual Meeting}} of the {{Association}} for {{Computational Linguistics}} ({{Volume}} 1: {{Long Papers}})},\n  author = {Lindemann, Matthias and Koller, Alexander and Titov, Ivan},\n  editor = {Rogers, Anna and {Boyd-Graber}, Jordan and Okazaki, Naoaki},\n  year = {2023},\n  month = jul,\n  pages = {14488--14506},\n  publisher = {Association for Computational Linguistics},\n  address = {Toronto, Canada},\n  doi = {10.18653/v1/2023.acl-long.810},\n  url = {https://aclanthology.org/2023.acl-long.810},\n  urldate = {2024-03-27},\n  abstract = {Seq2seq models have been shown to struggle with compositional generalization in semantic parsing, i.e. generalizing to unseen compositions of phenomena that the model handles correctly in isolation. We phrase semantic parsing as a two-step process: we first tag each input token with a multiset of output tokens. Then we arrange the tokens into an output sequence using a new way of parameterizing and predicting permutations. We formulate predicting a permutation as solving a regularized linear program and we backpropagate through the solver. In contrast to prior work, our approach does not place a priori restrictions on possible permutations, making it very expressive. Our model outperforms pretrained seq2seq models and prior work on realistic semantic parsing tasks that require generalization to longer examples. We also outperform non-tree-based models on structural generalization on the COGS benchmark. For the first time, we show that a model without an inductive bias provided by trees achieves high accuracy on generalization to deeper recursion depth.},\n  file = {/Users/shanest/Zotero/storage/54KQWV49/Lindemann et al. - 2023 - Compositional Generalization without Trees using M.pdf}\n}\n\n
\n
\n\n\n
\n Seq2seq models have been shown to struggle with compositional generalization in semantic parsing, i.e. generalizing to unseen compositions of phenomena that the model handles correctly in isolation. We phrase semantic parsing as a two-step process: we first tag each input token with a multiset of output tokens. Then we arrange the tokens into an output sequence using a new way of parameterizing and predicting permutations. We formulate predicting a permutation as solving a regularized linear program and we backpropagate through the solver. In contrast to prior work, our approach does not place a priori restrictions on possible permutations, making it very expressive. Our model outperforms pretrained seq2seq models and prior work on realistic semantic parsing tasks that require generalization to longer examples. We also outperform non-tree-based models on structural generalization on the COGS benchmark. For the first time, we show that a model without an inductive bias provided by trees achieves high accuracy on generalization to deeper recursion depth.\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n \n SLOG: A Structural Generalization Benchmark for Semantic Parsing.\n \n \n \n \n\n\n \n Li, B.; Donatelli, L.; Koller, A.; Linzen, T.; Yao, Y.; and Kim, N.\n\n\n \n\n\n\n In Bouamor, H.; Pino, J.; and Bali, K., editor(s), Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 3213–3232, Singapore, December 2023. Association for Computational Linguistics\n \n\n\n\n
\n\n\n\n \n \n \"SLOG:Paper\n  \n \n\n \n \n doi\n  \n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n  \n \n 2 downloads\n \n \n\n \n \n \n \n \n \n \n\n  \n \n \n\n\n\n
\n 
@inproceedings{liSLOGStructuralGeneralization2023,\n  title = {{{SLOG}}: {{A Structural Generalization Benchmark}} for {{Semantic Parsing}}},\n  shorttitle = {{{SLOG}}},\n  booktitle = {Proceedings of the 2023 {{Conference}} on {{Empirical Methods}} in {{Natural Language Processing}}},\n  author = {Li, Bingzhi and Donatelli, Lucia and Koller, Alexander and Linzen, Tal and Yao, Yuekun and Kim, Najoung},\n  editor = {Bouamor, Houda and Pino, Juan and Bali, Kalika},\n  year = {2023},\n  month = dec,\n  pages = {3213--3232},\n  publisher = {Association for Computational Linguistics},\n  address = {Singapore},\n  doi = {10.18653/v1/2023.emnlp-main.194},\n  url = {https://aclanthology.org/2023.emnlp-main.194},\n  urldate = {2024-03-18},\n  abstract = {The goal of compositional generalization benchmarks is to evaluate how well models generalize to new complex linguistic expressions. Existing benchmarks often focus on lexical generalization, the interpretation of novel lexical items in syntactic structures familiar from training; structural generalization tasks, where a model needs to interpret syntactic structures that are themselves unfamiliar from training, are often underrepresented, resulting in overly optimistic perceptions of how well models can generalize. We introduce SLOG, a semantic parsing dataset that extends COGS (Kim and Linzen, 2020) with 17 structural generalization cases. In our experiments, the generalization accuracy of Transformer models, including pretrained ones, only reaches 40.6\\%, while a structure-aware parser only achieves 70.8\\%. These results are far from the near-perfect accuracy existing models achieve on COGS, demonstrating the role of SLOG in foregrounding the large discrepancy between models' lexical and structural generalization capacities.},\n  file = {/Users/shanest/sync/library/Li et al/2023/Li et al. - 2023 - SLOG A Structural Generalization Benchmark for Se.pdf}\n}\n\n
\n
\n\n\n
\n The goal of compositional generalization benchmarks is to evaluate how well models generalize to new complex linguistic expressions. Existing benchmarks often focus on lexical generalization, the interpretation of novel lexical items in syntactic structures familiar from training; structural generalization tasks, where a model needs to interpret syntactic structures that are themselves unfamiliar from training, are often underrepresented, resulting in overly optimistic perceptions of how well models can generalize. We introduce SLOG, a semantic parsing dataset that extends COGS (Kim and Linzen, 2020) with 17 structural generalization cases. In our experiments, the generalization accuracy of Transformer models, including pretrained ones, only reaches 40.6%, while a structure-aware parser only achieves 70.8%. These results are far from the near-perfect accuracy existing models achieve on COGS, demonstrating the role of SLOG in foregrounding the large discrepancy between models' lexical and structural generalization capacities.\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n \n Measuring and Narrowing the Compositionality Gap in Language Models.\n \n \n \n \n\n\n \n Press, O.; Zhang, M.; Min, S.; Schmidt, L.; Smith, N.; and Lewis, M.\n\n\n \n\n\n\n In Bouamor, H.; Pino, J.; and Bali, K., editor(s), Findings of the Association for Computational Linguistics: EMNLP 2023, pages 5687–5711, Singapore, December 2023. Association for Computational Linguistics\n \n\n\n\n
\n\n\n\n \n \n \"MeasuringPaper\n  \n \n\n \n \n doi\n  \n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n  \n \n 1 download\n \n \n\n \n \n \n \n \n \n \n\n  \n \n \n\n\n\n
\n 
@inproceedings{pressMeasuringNarrowingCompositionality2023,\n  title = {Measuring and {{Narrowing}} the {{Compositionality Gap}} in {{Language Models}}},\n  booktitle = {Findings of the {{Association}} for {{Computational Linguistics}}: {{EMNLP}} 2023},\n  author = {Press, Ofir and Zhang, Muru and Min, Sewon and Schmidt, Ludwig and Smith, Noah and Lewis, Mike},\n  editor = {Bouamor, Houda and Pino, Juan and Bali, Kalika},\n  year = {2023},\n  month = dec,\n  pages = {5687--5711},\n  publisher = {Association for Computational Linguistics},\n  address = {Singapore},\n  doi = {10.18653/v1/2023.findings-emnlp.378},\n  url = {https://aclanthology.org/2023.findings-emnlp.378},\n  urldate = {2024-03-25},\n  abstract = {We investigate the ability of language models to perform compositional reasoning tasks where the overall solution depends on correctly composing the answers to sub-problems. We measure how often models can correctly answer all sub-problems but not generate the overall solution, a ratio we call the compositionality gap. We evaluate this ratio by asking multi-hop questions with answers that require composing multiple facts unlikely to have been observed together during pretraining. In the GPT-3 family of models, as model size increases we show that the single-hop question answering performance improves faster than the multi-hop performance does, therefore the compositionality gap does not decrease. This surprising result suggests that while more powerful models memorize and recall more factual knowledge, they show no corresponding improvement in their ability to perform this kind of compositional reasoning. We then demonstrate how elicitive prompting (such as chain of thought) narrows the compositionality gap by reasoning explicitly instead of implicitly. We present a new method, self-ask, that further improves on chain of thought. In our method, the model explicitly asks itself (and then answers) follow-up questions before answering the initial question. We finally show that self-ask's structured prompting lets us easily plug in a search engine to answer the follow-up questions, which additionally improves accuracy.},\n  file = {/Users/shanest/Zotero/storage/YJCMEHS6/Press et al. - 2023 - Measuring and Narrowing the Compositionality Gap i.pdf}\n}\n\n
\n
\n\n\n
\n We investigate the ability of language models to perform compositional reasoning tasks where the overall solution depends on correctly composing the answers to sub-problems. We measure how often models can correctly answer all sub-problems but not generate the overall solution, a ratio we call the compositionality gap. We evaluate this ratio by asking multi-hop questions with answers that require composing multiple facts unlikely to have been observed together during pretraining. In the GPT-3 family of models, as model size increases we show that the single-hop question answering performance improves faster than the multi-hop performance does, therefore the compositionality gap does not decrease. This surprising result suggests that while more powerful models memorize and recall more factual knowledge, they show no corresponding improvement in their ability to perform this kind of compositional reasoning. We then demonstrate how elicitive prompting (such as chain of thought) narrows the compositionality gap by reasoning explicitly instead of implicitly. We present a new method, self-ask, that further improves on chain of thought. In our method, the model explicitly asks itself (and then answers) follow-up questions before answering the initial question. We finally show that self-ask's structured prompting lets us easily plug in a search engine to answer the follow-up questions, which additionally improves accuracy.\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n \n mSCAN: A Dataset for Multilingual Compositional Generalisation Evaluation.\n \n \n \n \n\n\n \n Reymond, A.; and Steinert-Threlkeld, S.\n\n\n \n\n\n\n In Hupkes, D.; Dankers, V.; Batsuren, K.; Sinha, K.; Kazemnejad, A.; Christodoulopoulos, C.; Cotterell, R.; and Bruni, E., editor(s), Proceedings of the 1st GenBench Workshop on (Benchmarking) Generalisation in NLP, pages 143–151, Singapore, December 2023. Association for Computational Linguistics\n \n\n\n\n
\n\n\n\n \n \n \"mSCAN:Paper\n  \n \n\n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n  \n \n 4 downloads\n \n \n\n \n \n \n \n \n \n \n\n  \n \n \n\n\n\n
\n 
@inproceedings{reymondMSCANDatasetMultilingual2023,\n  title = {{{mSCAN}}: {{A Dataset}} for {{Multilingual Compositional Generalisation Evaluation}}},\n  shorttitle = {{{mSCAN}}},\n  booktitle = {Proceedings of the 1st {{GenBench Workshop}} on ({{Benchmarking}}) {{Generalisation}} in {{NLP}}},\n  author = {Reymond, Am{\\'e}lie and {Steinert-Threlkeld}, Shane},\n  editor = {Hupkes, Dieuwke and Dankers, Verna and Batsuren, Khuyagbaatar and Sinha, Koustuv and Kazemnejad, Amirhossein and Christodoulopoulos, Christos and Cotterell, Ryan and Bruni, Elia},\n  year = {2023},\n  month = dec,\n  pages = {143--151},\n  publisher = {Association for Computational Linguistics},\n  address = {Singapore},\n  url = {https://aclanthology.org/2023.genbench-1.11},\n  urldate = {2023-12-14},\n  abstract = {Language models achieve remarkable results on a variety of tasks, yet still struggle on compositional generalisation benchmarks. The majority of these benchmarks evaluate performance in English only, leaving us with the question of whether these results generalise to other languages. As an initial step to answering this question, we introduce mSCAN, a multilingual adaptation of the SCAN dataset. It was produced by a rule-based translation, developed in cooperation with native speakers. We then showcase this novel dataset on some in-context learning experiments, and GPT3.5 and the multilingual large language model BLOOM},\n  file = {/Users/shanest/sync/library/Reymond_Steinert-Threlkeld/2023/Reymond and Steinert-Threlkeld - 2023 - mSCAN A Dataset for Multilingual Compositional Ge.pdf}\n}\n\n
\n
\n\n\n
\n Language models achieve remarkable results on a variety of tasks, yet still struggle on compositional generalisation benchmarks. The majority of these benchmarks evaluate performance in English only, leaving us with the question of whether these results generalise to other languages. As an initial step to answering this question, we introduce mSCAN, a multilingual adaptation of the SCAN dataset. It was produced by a rule-based translation, developed in cooperation with native speakers. We then showcase this novel dataset on some in-context learning experiments, and GPT3.5 and the multilingual large language model BLOOM\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n \n Testing the General Deductive Reasoning Capacity of Large Language Models Using OOD Examples.\n \n \n \n \n\n\n \n Saparov, A.; Pang, R. Y.; Padmakumar, V.; Joshi, N.; Kazemi, M.; Kim, N.; and He, H.\n\n\n \n\n\n\n In Thirty-Seventh Conference on Neural Information Processing Systems, November 2023. \n \n\n\n\n
\n\n\n\n \n \n \"TestingPaper\n  \n \n\n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n\n\n\n
\n 
@inproceedings{saparovTestingGeneralDeductive2023,\n  title = {Testing the {{General Deductive Reasoning Capacity}} of {{Large Language Models Using OOD Examples}}},\n  booktitle = {Thirty-Seventh {{Conference}} on {{Neural Information Processing Systems}}},\n  author = {Saparov, Abulhair and Pang, Richard Yuanzhe and Padmakumar, Vishakh and Joshi, Nitish and Kazemi, Mehran and Kim, Najoung and He, He},\n  year = {2023},\n  month = nov,\n  url = {https://openreview.net/forum?id=MCVfX7HgPO&referrer=%5Bthe%20profile%20of%20He%20He%5D(%2Fprofile%3Fid%3D~He_He2)},\n  urldate = {2024-03-19},\n  abstract = {Given the intractably large size of the space of proofs, any model that is capable of general deductive reasoning must generalize to proofs of greater complexity. Recent studies have shown that large language models (LLMs) possess some abstract deductive reasoning ability given chain-of-thought prompts. However, they have primarily been tested on proofs using modus ponens or of a specific size, and from the same distribution as the in-context examples. To measure the general deductive reasoning ability of LLMs, we test on a broad set of deduction rules and measure their ability to generalize to more complex proofs from simpler demonstrations from multiple angles: depth-, width-, and compositional generalization. To facilitate systematic exploration, we construct a new synthetic and programmable reasoning dataset that enables control over deduction rules and proof complexity. Our experiments on four LLMs of various sizes and training objectives show that they are able to generalize to compositional proofs. However, they have difficulty generalizing to longer proofs, and they require explicit demonstrations to produce hypothetical subproofs, specifically in proof by cases and proof by contradiction.},\n  langid = {english},\n  file = {/Users/shanest/sync/library/Saparov et al/2023/Saparov et al. - 2023 - Testing the General Deductive Reasoning Capacity o.pdf}\n}\n\n
\n
\n\n\n
\n Given the intractably large size of the space of proofs, any model that is capable of general deductive reasoning must generalize to proofs of greater complexity. Recent studies have shown that large language models (LLMs) possess some abstract deductive reasoning ability given chain-of-thought prompts. However, they have primarily been tested on proofs using modus ponens or of a specific size, and from the same distribution as the in-context examples. To measure the general deductive reasoning ability of LLMs, we test on a broad set of deduction rules and measure their ability to generalize to more complex proofs from simpler demonstrations from multiple angles: depth-, width-, and compositional generalization. To facilitate systematic exploration, we construct a new synthetic and programmable reasoning dataset that enables control over deduction rules and proof complexity. Our experiments on four LLMs of various sizes and training objectives show that they are able to generalize to compositional proofs. However, they have difficulty generalizing to longer proofs, and they require explicit demonstrations to produce hypothetical subproofs, specifically in proof by cases and proof by contradiction.\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n \n Least-to-Most Prompting Enables Complex Reasoning in Large Language Models.\n \n \n \n \n\n\n \n Zhou, D.; Schärli, N.; Hou, L.; Wei, J.; Scales, N.; Wang, X.; Schuurmans, D.; Cui, C.; Bousquet, O.; Le, Q. V.; and Chi, E. H.\n\n\n \n\n\n\n In The Eleventh International Conference on Learning Representations, 2023. \n \n\n\n\n
\n\n\n\n \n \n \"Least-to-MostPaper\n  \n \n\n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n  \n \n 1 download\n \n \n\n \n \n \n \n \n \n \n\n  \n \n \n\n\n\n
\n 
@inproceedings{zhouLeasttoMostPromptingEnables2023,\n  title = {Least-to-{{Most Prompting Enables Complex Reasoning}} in {{Large Language Models}}},\n  booktitle = {The {{Eleventh International Conference}} on {{Learning Representations}}},\n  author = {Zhou, Denny and Sch{\\"a}rli, Nathanael and Hou, Le and Wei, Jason and Scales, Nathan and Wang, Xuezhi and Schuurmans, Dale and Cui, Claire and Bousquet, Olivier and Le, Quoc V. and Chi, Ed H.},\n  year = {2023},\n  url = {https://openreview.net/forum?id=WZH7099tgfM},\n  urldate = {2024-03-19},\n  abstract = {Chain-of-thought prompting has demonstrated remarkable performance on various natural language reasoning tasks. However, it tends to perform poorly on tasks which requires solving problems harder than the exemplars shown in the prompts. To overcome this challenge of easy-to-hard generalization, we propose a novel prompting strategy, least-to-most prompting. The key idea in this strategy is to break down a complex problem into a series of simpler subproblems and then solve them in sequence. Solving each subproblem is facilitated by the answers to previously solved subproblems. Our experimental results on tasks related to symbolic manipulation, compositional generalization, and math reasoning reveal that least-to-most prompting is capable of generalizing to more difficult problems than those seen in the prompts. A notable finding is that when the GPT-3 code-davinci-002 model is used with least-to-most prompting, it can solve the compositional generalization benchmark SCAN in any split (including length split) with an accuracy of at least 99{\\textbackslash}\\% using just 14 exemplars, compared to only 16{\\textbackslash}\\% accuracy with chain-of-thought prompting. This is particularly noteworthy because neural-symbolic models in the literature that specialize in solving SCAN are trained on the entire training set containing over 15,000 examples. We have included prompts for all the tasks in the Appendix.},\n  langid = {english},\n  file = {/Users/shanest/sync/library/Zhou et al/2022/Zhou et al. - 2022 - Least-to-Most Prompting Enables Complex Reasoning .pdf}\n}\n\n
\n
\n\n\n
\n Chain-of-thought prompting has demonstrated remarkable performance on various natural language reasoning tasks. However, it tends to perform poorly on tasks which requires solving problems harder than the exemplars shown in the prompts. To overcome this challenge of easy-to-hard generalization, we propose a novel prompting strategy, least-to-most prompting. The key idea in this strategy is to break down a complex problem into a series of simpler subproblems and then solve them in sequence. Solving each subproblem is facilitated by the answers to previously solved subproblems. Our experimental results on tasks related to symbolic manipulation, compositional generalization, and math reasoning reveal that least-to-most prompting is capable of generalizing to more difficult problems than those seen in the prompts. A notable finding is that when the GPT-3 code-davinci-002 model is used with least-to-most prompting, it can solve the compositional generalization benchmark SCAN in any split (including length split) with an accuracy of at least 99\\% using just 14 exemplars, compared to only 16\\% accuracy with chain-of-thought prompting. This is particularly noteworthy because neural-symbolic models in the literature that specialize in solving SCAN are trained on the entire training set containing over 15,000 examples. We have included prompts for all the tasks in the Appendix.\n
\n\n\n
\n\n\n\n\n\n
\n
\n\n
\n
\n  \n 2022\n \n \n (3)\n \n \n
\n
\n \n \n
\n \n\n \n \n \n \n \n \n On the Compositional Generalization Gap of In-Context Learning.\n \n \n \n \n\n\n \n Hosseini, A.; Vani, A.; Bahdanau, D.; Sordoni, A.; and Courville, A.\n\n\n \n\n\n\n In Bastings, J.; Belinkov, Y.; Elazar, Y.; Hupkes, D.; Saphra, N.; and Wiegreffe, S., editor(s), Proceedings of the Fifth BlackboxNLP Workshop on Analyzing and Interpreting Neural Networks for NLP, pages 272–280, Abu Dhabi, United Arab Emirates (Hybrid), December 2022. Association for Computational Linguistics\n \n\n\n\n
\n\n\n\n \n \n \"OnPaper\n  \n \n\n \n \n doi\n  \n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n  \n \n 2 downloads\n \n \n\n \n \n \n \n \n \n \n\n  \n \n \n\n\n\n
\n 
@inproceedings{hosseiniCompositionalGeneralizationGap2022,\n  title = {On the {{Compositional Generalization Gap}} of {{In-Context Learning}}},\n  booktitle = {Proceedings of the {{Fifth BlackboxNLP Workshop}} on {{Analyzing}} and {{Interpreting Neural Networks}} for {{NLP}}},\n  author = {Hosseini, Arian and Vani, Ankit and Bahdanau, Dzmitry and Sordoni, Alessandro and Courville, Aaron},\n  editor = {Bastings, Jasmijn and Belinkov, Yonatan and Elazar, Yanai and Hupkes, Dieuwke and Saphra, Naomi and Wiegreffe, Sarah},\n  year = {2022},\n  month = dec,\n  pages = {272--280},\n  publisher = {Association for Computational Linguistics},\n  address = {Abu Dhabi, United Arab Emirates (Hybrid)},\n  doi = {10.18653/v1/2022.blackboxnlp-1.22},\n  url = {https://aclanthology.org/2022.blackboxnlp-1.22},\n  urldate = {2024-03-19},\n  abstract = {Pretrained large generative language models have shown great performance on many tasks, but exhibit low compositional generalization abilities. Scaling such models has been shown to improve their performance on various NLP tasks even just by conditioning them on a few examples to solve the task without any fine-tuning (also known as in-context learning). In this work, we look at the gap between the in-distribution (ID) and out-of-distribution (OOD) performance of such models in semantic parsing tasks with in-context learning. In the ID settings, the demonstrations are from the same split (test or train) that the model is being evaluated on, and in the OOD settings, they are from the other split. We look at how the relative generalization gap of in-context learning evolves as models are scaled up. We evaluate four model families, OPT, BLOOM, CodeGen and Codex on three semantic parsing datasets, CFQ, SCAN and GeoQuery with different number of exemplars, and observe a trend of decreasing relative generalization gap as models are scaled up.},\n  file = {/Users/shanest/sync/library/Hosseini et al/2022/Hosseini et al. - 2022 - On the Compositional Generalization Gap of In-Cont.pdf}\n}\n\n
\n
\n\n\n
\n Pretrained large generative language models have shown great performance on many tasks, but exhibit low compositional generalization abilities. Scaling such models has been shown to improve their performance on various NLP tasks even just by conditioning them on a few examples to solve the task without any fine-tuning (also known as in-context learning). In this work, we look at the gap between the in-distribution (ID) and out-of-distribution (OOD) performance of such models in semantic parsing tasks with in-context learning. In the ID settings, the demonstrations are from the same split (test or train) that the model is being evaluated on, and in the OOD settings, they are from the other split. We look at how the relative generalization gap of in-context learning evolves as models are scaled up. We evaluate four model families, OPT, BLOOM, CodeGen and Codex on three semantic parsing datasets, CFQ, SCAN and GeoQuery with different number of exemplars, and observe a trend of decreasing relative generalization gap as models are scaled up.\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n \n Making Transformers Solve Compositional Tasks.\n \n \n \n \n\n\n \n Ontanon, S.; Ainslie, J.; Fisher, Z.; and Cvicek, V.\n\n\n \n\n\n\n In Muresan, S.; Nakov, P.; and Villavicencio, A., editor(s), Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 3591–3607, Dublin, Ireland, May 2022. Association for Computational Linguistics\n \n\n\n\n
\n\n\n\n \n \n \"MakingPaper\n  \n \n\n \n \n doi\n  \n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n\n\n\n
\n 
@inproceedings{ontanonMakingTransformersSolve2022,\n  title = {Making {{Transformers Solve Compositional Tasks}}},\n  booktitle = {Proceedings of the 60th {{Annual Meeting}} of the {{Association}} for {{Computational Linguistics}} ({{Volume}} 1: {{Long Papers}})},\n  author = {Ontanon, Santiago and Ainslie, Joshua and Fisher, Zachary and Cvicek, Vaclav},\n  editor = {Muresan, Smaranda and Nakov, Preslav and Villavicencio, Aline},\n  year = {2022},\n  month = may,\n  pages = {3591--3607},\n  publisher = {Association for Computational Linguistics},\n  address = {Dublin, Ireland},\n  doi = {10.18653/v1/2022.acl-long.251},\n  url = {https://aclanthology.org/2022.acl-long.251},\n  urldate = {2024-03-19},\n  abstract = {Several studies have reported the inability of Transformer models to generalize compositionally, a key type of generalization in many NLP tasks such as semantic parsing. In this paper we explore the design space of Transformer models showing that the inductive biases given to the model by several design decisions significantly impact compositional generalization. We identified Transformer configurations that generalize compositionally significantly better than previously reported in the literature in many compositional tasks. We achieve state-of-the-art results in a semantic parsing compositional generalization benchmark (COGS), and a string edit operation composition benchmark (PCFG).},\n  file = {/Users/shanest/sync/library/Ontanon et al/2022/Ontanon et al. - 2022 - Making Transformers Solve Compositional Tasks.pdf}\n}\n\n
\n
\n\n\n
\n Several studies have reported the inability of Transformer models to generalize compositionally, a key type of generalization in many NLP tasks such as semantic parsing. In this paper we explore the design space of Transformer models showing that the inductive biases given to the model by several design decisions significantly impact compositional generalization. We identified Transformer configurations that generalize compositionally significantly better than previously reported in the literature in many compositional tasks. We achieve state-of-the-art results in a semantic parsing compositional generalization benchmark (COGS), and a string edit operation composition benchmark (PCFG).\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n \n Compositionality.\n \n \n \n \n\n\n \n Szabó, Z. G.\n\n\n \n\n\n\n The Stanford Encyclopedia of Philosophy, Fall. 2022.\n \n\n\n\n
\n\n\n\n \n \n \"CompositionalityPaper\n  \n \n\n \n\n \n link\n  \n \n\n bibtex\n \n\n \n\n \n  \n \n 6 downloads\n \n \n\n \n \n \n \n \n \n \n\n  \n \n \n\n\n\n
\n 
@article{szaboCompositionality2022,\n  title = {Compositionality},\n  author = {Szab{\\'o}, Zolt{\\'a}n Gendler},\n  editor = {Zalta, Edward N},\n  year = {2022},\n  journal = {The Stanford Encyclopedia of Philosophy},\n  volume = {Fall},\n  url = {https://plato.stanford.edu/archives/fall2022/entries/compositionality/}\n}\n\n
\n
\n\n\n\n
\n\n\n\n\n\n
\n
\n\n
\n
\n  \n 2021\n \n \n (8)\n \n \n
\n
\n \n \n
\n \n\n \n \n \n \n \n \n Can Transformers Jump Around Right in Natural Language? Assessing Performance Transfer from SCAN.\n \n \n \n \n\n\n \n Chaabouni, R.; Dessì, Roberto; and Kharitonov, E.\n\n\n \n\n\n\n In Bastings, J.; Belinkov, Y.; Dupoux, E.; Giulianelli, M.; Hupkes, D.; Pinter, Y.; and Sajjad, H., editor(s), Proceedings of the Fourth BlackboxNLP Workshop on Analyzing and Interpreting Neural Networks for NLP, pages 136–148, Punta Cana, Dominican Republic, November 2021. Association for Computational Linguistics\n \n\n\n\n
\n\n\n\n \n \n \"CanPaper\n  \n \n\n \n \n doi\n  \n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n\n\n\n
\n 
@inproceedings{chaabouniCanTransformersJump2021,\n  title = {Can {{Transformers Jump Around Right}} in {{Natural Language}}? {{Assessing Performance Transfer}} from {{SCAN}}},\n  shorttitle = {Can {{Transformers Jump Around Right}} in {{Natural Language}}?},\n  booktitle = {Proceedings of the {{Fourth BlackboxNLP Workshop}} on {{Analyzing}} and {{Interpreting Neural Networks}} for {{NLP}}},\n  author = {Chaabouni, Rahma and Dess{\\`i}, Roberto and Kharitonov, Eugene},\n  editor = {Bastings, Jasmijn and Belinkov, Yonatan and Dupoux, Emmanuel and Giulianelli, Mario and Hupkes, Dieuwke and Pinter, Yuval and Sajjad, Hassan},\n  year = {2021},\n  month = nov,\n  pages = {136--148},\n  publisher = {Association for Computational Linguistics},\n  address = {Punta Cana, Dominican Republic},\n  doi = {10.18653/v1/2021.blackboxnlp-1.9},\n  url = {https://aclanthology.org/2021.blackboxnlp-1.9},\n  urldate = {2024-03-19},\n  abstract = {Despite their failure to solve the compositional SCAN dataset, seq2seq architectures still achieve astonishing success on more practical tasks. This observation pushes us to question the usefulness of SCAN-style compositional generalization in realistic NLP tasks. In this work, we study the benefit that such compositionality brings about to several machine translation tasks. We present several focused modifications of Transformer that greatly improve generalization capabilities on SCAN and select one that remains on par with a vanilla Transformer on a standard machine translation (MT) task. Next, we study its performance in low-resource settings and on a newly introduced distribution-shifted English-French translation task. Overall, we find that improvements of a SCAN-capable model do not directly transfer to the resource-rich MT setup. In contrast, in the low-resource setup, general modifications lead to an improvement of up to 13.1\\% BLEU score w.r.t. a vanilla Transformer. Similarly, an improvement of 14\\% in an accuracy-based metric is achieved in the introduced compositional English-French translation task. This provides experimental evidence that the compositional generalization assessed in SCAN is particularly useful in resource-starved and domain-shifted scenarios.},\n  file = {/Users/shanest/sync/library/Chaabouni et al/2021/Chaabouni et al. - 2021 - Can Transformers Jump Around Right in Natural Lang.pdf}\n}\n\n
\n
\n\n\n
\n Despite their failure to solve the compositional SCAN dataset, seq2seq architectures still achieve astonishing success on more practical tasks. This observation pushes us to question the usefulness of SCAN-style compositional generalization in realistic NLP tasks. In this work, we study the benefit that such compositionality brings about to several machine translation tasks. We present several focused modifications of Transformer that greatly improve generalization capabilities on SCAN and select one that remains on par with a vanilla Transformer on a standard machine translation (MT) task. Next, we study its performance in low-resource settings and on a newly introduced distribution-shifted English-French translation task. Overall, we find that improvements of a SCAN-capable model do not directly transfer to the resource-rich MT setup. In contrast, in the low-resource setup, general modifications lead to an improvement of up to 13.1% BLEU score w.r.t. a vanilla Transformer. Similarly, an improvement of 14% in an accuracy-based metric is achieved in the introduced compositional English-French translation task. This provides experimental evidence that the compositional generalization assessed in SCAN is particularly useful in resource-starved and domain-shifted scenarios.\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n \n Meta-Learning to Compositionally Generalize.\n \n \n \n \n\n\n \n Conklin, H.; Wang, B.; Smith, K.; and Titov, I.\n\n\n \n\n\n\n In Zong, C.; Xia, F.; Li, W.; and Navigli, R., editor(s), Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers), pages 3322–3335, Online, August 2021. Association for Computational Linguistics\n \n\n\n\n
\n\n\n\n \n \n \"Meta-LearningPaper\n  \n \n\n \n \n doi\n  \n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n\n\n\n
\n 
@inproceedings{conklinMetaLearningCompositionallyGeneralize2021,\n  title = {Meta-{{Learning}} to {{Compositionally Generalize}}},\n  booktitle = {Proceedings of the 59th {{Annual Meeting}} of the {{Association}} for {{Computational Linguistics}} and the 11th {{International Joint Conference}} on {{Natural Language Processing}} ({{Volume}} 1: {{Long Papers}})},\n  author = {Conklin, Henry and Wang, Bailin and Smith, Kenny and Titov, Ivan},\n  editor = {Zong, Chengqing and Xia, Fei and Li, Wenjie and Navigli, Roberto},\n  year = {2021},\n  month = aug,\n  pages = {3322--3335},\n  publisher = {Association for Computational Linguistics},\n  address = {Online},\n  doi = {10.18653/v1/2021.acl-long.258},\n  url = {https://aclanthology.org/2021.acl-long.258},\n  urldate = {2024-03-19},\n  abstract = {Natural language is compositional; the meaning of a sentence is a function of the meaning of its parts. This property allows humans to create and interpret novel sentences, generalizing robustly outside their prior experience. Neural networks have been shown to struggle with this kind of generalization, in particular performing poorly on tasks designed to assess compositional generalization (i.e. where training and testing distributions differ in ways that would be trivial for a compositional strategy to resolve). Their poor performance on these tasks may in part be due to the nature of supervised learning which assumes training and testing data to be drawn from the same distribution. We implement a meta-learning augmented version of supervised learning whose objective directly optimizes for out-of-distribution generalization. We construct pairs of tasks for meta-learning by sub-sampling existing training data. Each pair of tasks is constructed to contain relevant examples, as determined by a similarity metric, in an effort to inhibit models from memorizing their input. Experimental results on the COGS and SCAN datasets show that our similarity-driven meta-learning can improve generalization performance.},\n  file = {/Users/shanest/sync/library/Conklin et al/2021/Conklin et al. - 2021 - Meta-Learning to Compositionally Generalize.pdf}\n}\n\n
\n
\n\n\n
\n Natural language is compositional; the meaning of a sentence is a function of the meaning of its parts. This property allows humans to create and interpret novel sentences, generalizing robustly outside their prior experience. Neural networks have been shown to struggle with this kind of generalization, in particular performing poorly on tasks designed to assess compositional generalization (i.e. where training and testing distributions differ in ways that would be trivial for a compositional strategy to resolve). Their poor performance on these tasks may in part be due to the nature of supervised learning which assumes training and testing data to be drawn from the same distribution. We implement a meta-learning augmented version of supervised learning whose objective directly optimizes for out-of-distribution generalization. We construct pairs of tasks for meta-learning by sub-sampling existing training data. Each pair of tasks is constructed to contain relevant examples, as determined by a similarity metric, in an effort to inhibit models from memorizing their input. Experimental results on the COGS and SCAN datasets show that our similarity-driven meta-learning can improve generalization performance.\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n \n The Devil Is in the Detail: Simple Tricks Improve Systematic Generalization of Transformers.\n \n \n \n \n\n\n \n Csordás, R.; Irie, K.; and Schmidhuber, J.\n\n\n \n\n\n\n In Moens, M.; Huang, X.; Specia, L.; and Yih, S. W., editor(s), Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 619–634, Online and Punta Cana, Dominican Republic, November 2021. Association for Computational Linguistics\n \n\n\n\n
\n\n\n\n \n \n \"ThePaper\n  \n \n\n \n \n doi\n  \n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n\n\n\n
\n 
@inproceedings{csordasDevilDetailSimple2021,\n  title = {The {{Devil}} Is in the {{Detail}}: {{Simple Tricks Improve Systematic Generalization}} of {{Transformers}}},\n  shorttitle = {The {{Devil}} Is in the {{Detail}}},\n  booktitle = {Proceedings of the 2021 {{Conference}} on {{Empirical Methods}} in {{Natural Language Processing}}},\n  author = {Csord{\\'a}s, R{\\'o}bert and Irie, Kazuki and Schmidhuber, Juergen},\n  editor = {Moens, Marie-Francine and Huang, Xuanjing and Specia, Lucia and Yih, Scott Wen-tau},\n  year = {2021},\n  month = nov,\n  pages = {619--634},\n  publisher = {Association for Computational Linguistics},\n  address = {Online and Punta Cana, Dominican Republic},\n  doi = {10.18653/v1/2021.emnlp-main.49},\n  url = {https://aclanthology.org/2021.emnlp-main.49},\n  urldate = {2024-03-19},\n  abstract = {Recently, many datasets have been proposed to test the systematic generalization ability of neural networks. The companion baseline Transformers, typically trained with default hyper-parameters from standard tasks, are shown to fail dramatically. Here we demonstrate that by revisiting model configurations as basic as scaling of embeddings, early stopping, relative positional embedding, and Universal Transformer variants, we can drastically improve the performance of Transformers on systematic generalization. We report improvements on five popular datasets: SCAN, CFQ, PCFG, COGS, and Mathematics dataset. Our models improve accuracy from 50\\% to 85\\% on the PCFG productivity split, and from 35\\% to 81\\% on COGS. On SCAN, relative positional embedding largely mitigates the EOS decision problem (Newman et al., 2020), yielding 100\\% accuracy on the length split with a cutoff at 26. Importantly, performance differences between these models are typically invisible on the IID data split. This calls for proper generalization validation sets for developing neural networks that generalize systematically. We publicly release the code to reproduce our results.},\n  file = {/Users/shanest/sync/library/Csordás et al/2021/Csordás et al. - 2021 - The Devil is in the Detail Simple Tricks Improve .pdf}\n}\n\n
\n
\n\n\n
\n Recently, many datasets have been proposed to test the systematic generalization ability of neural networks. The companion baseline Transformers, typically trained with default hyper-parameters from standard tasks, are shown to fail dramatically. Here we demonstrate that by revisiting model configurations as basic as scaling of embeddings, early stopping, relative positional embedding, and Universal Transformer variants, we can drastically improve the performance of Transformers on systematic generalization. We report improvements on five popular datasets: SCAN, CFQ, PCFG, COGS, and Mathematics dataset. Our models improve accuracy from 50% to 85% on the PCFG productivity split, and from 35% to 81% on COGS. On SCAN, relative positional embedding largely mitigates the EOS decision problem (Newman et al., 2020), yielding 100% accuracy on the length split with a cutoff at 26. Importantly, performance differences between these models are typically invisible on the IID data split. This calls for proper generalization validation sets for developing neural networks that generalize systematically. We publicly release the code to reproduce our results.\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n \n Span-Based Semantic Parsing for Compositional Generalization.\n \n \n \n \n\n\n \n Herzig, J.; and Berant, J.\n\n\n \n\n\n\n In Zong, C.; Xia, F.; Li, W.; and Navigli, R., editor(s), Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers), pages 908–921, Online, August 2021. Association for Computational Linguistics\n \n\n\n\n
\n\n\n\n \n \n \"Span-BasedPaper\n  \n \n\n \n \n doi\n  \n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n  \n \n 11 downloads\n \n \n\n \n \n \n \n \n \n \n\n  \n \n \n\n\n\n
\n 
@inproceedings{herzigSpanbasedSemanticParsing2021,\n  title = {Span-Based {{Semantic Parsing}} for {{Compositional Generalization}}},\n  booktitle = {Proceedings of the 59th {{Annual Meeting}} of the {{Association}} for {{Computational Linguistics}} and the 11th {{International Joint Conference}} on {{Natural Language Processing}} ({{Volume}} 1: {{Long Papers}})},\n  author = {Herzig, Jonathan and Berant, Jonathan},\n  editor = {Zong, Chengqing and Xia, Fei and Li, Wenjie and Navigli, Roberto},\n  year = {2021},\n  month = aug,\n  pages = {908--921},\n  publisher = {Association for Computational Linguistics},\n  address = {Online},\n  doi = {10.18653/v1/2021.acl-long.74},\n  url = {https://aclanthology.org/2021.acl-long.74},\n  urldate = {2024-03-19},\n  abstract = {Despite the success of sequence-to-sequence (seq2seq) models in semantic parsing, recent work has shown that they fail in compositional generalization, i.e., the ability to generalize to new structures built of components observed during training. In this work, we posit that a span-based parser should lead to better compositional generalization. we propose SpanBasedSP, a parser that predicts a span tree over an input utterance, explicitly encoding how partial programs compose over spans in the input. SpanBasedSP extends Pasupat et al. (2019) to be comparable to seq2seq models by (i) training from programs, without access to gold trees, treating trees as latent variables, (ii) parsing a class of non-projective trees through an extension to standard CKY. On GeoQuery, SCAN and CLOSURE datasets, SpanBasedSP performs similarly to strong seq2seq baselines on random splits, but dramatically improves performance compared to baselines on splits that require compositional generalization: from 61.0 {$\\rightarrow$} 88.9 average accuracy.},\n  file = {/Users/shanest/sync/library/Herzig_Berant/2021/Herzig and Berant - 2021 - Span-based Semantic Parsing for Compositional Gene.pdf}\n}\n\n
\n
\n\n\n
\n Despite the success of sequence-to-sequence (seq2seq) models in semantic parsing, recent work has shown that they fail in compositional generalization, i.e., the ability to generalize to new structures built of components observed during training. In this work, we posit that a span-based parser should lead to better compositional generalization. we propose SpanBasedSP, a parser that predicts a span tree over an input utterance, explicitly encoding how partial programs compose over spans in the input. SpanBasedSP extends Pasupat et al. (2019) to be comparable to seq2seq models by (i) training from programs, without access to gold trees, treating trees as latent variables, (ii) parsing a class of non-projective trees through an extension to standard CKY. On GeoQuery, SCAN and CLOSURE datasets, SpanBasedSP performs similarly to strong seq2seq baselines on random splits, but dramatically improves performance compared to baselines on splits that require compositional generalization: from 61.0 $→$ 88.9 average accuracy.\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n \n Sequence-to-Sequence Learning with Latent Neural Grammars.\n \n \n \n \n\n\n \n Kim, Y.\n\n\n \n\n\n\n In Advances in Neural Information Processing Systems, volume 34, pages 26302–26317, 2021. Curran Associates, Inc.\n \n\n\n\n
\n\n\n\n \n \n \"Sequence-to-SequencePaper\n  \n \n\n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n\n\n\n
\n 
@inproceedings{kimSequencetoSequenceLearningLatent2021,\n  title = {Sequence-to-{{Sequence Learning}} with {{Latent Neural Grammars}}},\n  booktitle = {Advances in {{Neural Information Processing Systems}}},\n  author = {Kim, Yoon},\n  year = {2021},\n  volume = {34},\n  pages = {26302--26317},\n  publisher = {Curran Associates, Inc.},\n  url = {https://papers.nips.cc/paper/2021/hash/dd17e652cd2a08fdb8bf7f68e2ad3814-Abstract.html},\n  urldate = {2024-04-02},\n  abstract = {Sequence-to-sequence learning with neural networks has become the de facto standard for sequence modeling. This approach typically models the local distribution over the next element with a powerful neural network that can condition on arbitrary context. While flexible and performant, these models often require large datasets for training and can fail spectacularly on benchmarks designed to test for compositional generalization. This work explores an alternative, hierarchical approach to sequence-to-sequence learning with synchronous grammars, where each node in the target tree is transduced by a subset of nodes in the source tree. The source and target trees are treated as fully latent and marginalized out during training. We develop a neural parameterization of the grammar which enables parameter sharing over combinatorial structures without the need for manual feature engineering. We apply this latent neural grammar to various domains---a diagnostic language navigation task designed to test for compositional generalization (SCAN), style transfer, and small-scale machine translation---and find that it performs respectably compared to standard baselines.}\n}\n\n
\n
\n\n\n
\n Sequence-to-sequence learning with neural networks has become the de facto standard for sequence modeling. This approach typically models the local distribution over the next element with a powerful neural network that can condition on arbitrary context. While flexible and performant, these models often require large datasets for training and can fail spectacularly on benchmarks designed to test for compositional generalization. This work explores an alternative, hierarchical approach to sequence-to-sequence learning with synchronous grammars, where each node in the target tree is transduced by a subset of nodes in the source tree. The source and target trees are treated as fully latent and marginalized out during training. We develop a neural parameterization of the grammar which enables parameter sharing over combinatorial structures without the need for manual feature engineering. We apply this latent neural grammar to various domains—a diagnostic language navigation task designed to test for compositional generalization (SCAN), style transfer, and small-scale machine translation—and find that it performs respectably compared to standard baselines.\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n \n On Compositional Generalization of Neural Machine Translation.\n \n \n \n \n\n\n \n Li, Y.; Yin, Y.; Chen, Y.; and Zhang, Y.\n\n\n \n\n\n\n In Zong, C.; Xia, F.; Li, W.; and Navigli, R., editor(s), Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers), pages 4767–4780, Online, August 2021. Association for Computational Linguistics\n \n\n\n\n
\n\n\n\n \n \n \"OnPaper\n  \n \n\n \n \n doi\n  \n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n\n\n\n
\n 
@inproceedings{liCompositionalGeneralizationNeural2021,\n  title = {On {{Compositional Generalization}} of {{Neural Machine Translation}}},\n  booktitle = {Proceedings of the 59th {{Annual Meeting}} of the {{Association}} for {{Computational Linguistics}} and the 11th {{International Joint Conference}} on {{Natural Language Processing}} ({{Volume}} 1: {{Long Papers}})},\n  author = {Li, Yafu and Yin, Yongjing and Chen, Yulong and Zhang, Yue},\n  editor = {Zong, Chengqing and Xia, Fei and Li, Wenjie and Navigli, Roberto},\n  year = {2021},\n  month = aug,\n  pages = {4767--4780},\n  publisher = {Association for Computational Linguistics},\n  address = {Online},\n  doi = {10.18653/v1/2021.acl-long.368},\n  url = {https://aclanthology.org/2021.acl-long.368},\n  urldate = {2024-03-19},\n  abstract = {Modern neural machine translation (NMT) models have achieved competitive performance in standard benchmarks such as WMT. However, there still exist significant issues such as robustness, domain generalization, etc. In this paper, we study NMT models from the perspective of compositional generalization by building a benchmark dataset, CoGnition, consisting of 216k clean and consistent sentence pairs. We quantitatively analyze effects of various factors using compound translation error rate, then demonstrate that the NMT model fails badly on compositional generalization, although it performs remarkably well under traditional metrics.},\n  file = {/Users/shanest/sync/library/Li et al/2021/Li et al. - 2021 - On Compositional Generalization of Neural Machine .pdf}\n}\n\n
\n
\n\n\n
\n Modern neural machine translation (NMT) models have achieved competitive performance in standard benchmarks such as WMT. However, there still exist significant issues such as robustness, domain generalization, etc. In this paper, we study NMT models from the perspective of compositional generalization by building a benchmark dataset, CoGnition, consisting of 216k clean and consistent sentence pairs. We quantitatively analyze effects of various factors using compound translation error rate, then demonstrate that the NMT model fails badly on compositional generalization, although it performs remarkably well under traditional metrics.\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n \n Compositional Generalization and Natural Language Variation: Can a Semantic Parsing Approach Handle Both?.\n \n \n \n \n\n\n \n Shaw, P.; Chang, M.; Pasupat, P.; and Toutanova, K.\n\n\n \n\n\n\n In Zong, C.; Xia, F.; Li, W.; and Navigli, R., editor(s), Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers), pages 922–938, Online, August 2021. Association for Computational Linguistics\n \n\n\n\n
\n\n\n\n \n \n \"CompositionalPaper\n  \n \n\n \n \n doi\n  \n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n\n\n\n
\n 
@inproceedings{shawCompositionalGeneralizationNatural2021,\n  title = {Compositional {{Generalization}} and {{Natural Language Variation}}: {{Can}} a {{Semantic Parsing Approach Handle Both}}?},\n  shorttitle = {Compositional {{Generalization}} and {{Natural Language Variation}}},\n  booktitle = {Proceedings of the 59th {{Annual Meeting}} of the {{Association}} for {{Computational Linguistics}} and the 11th {{International Joint Conference}} on {{Natural Language Processing}} ({{Volume}} 1: {{Long Papers}})},\n  author = {Shaw, Peter and Chang, Ming-Wei and Pasupat, Panupong and Toutanova, Kristina},\n  editor = {Zong, Chengqing and Xia, Fei and Li, Wenjie and Navigli, Roberto},\n  year = {2021},\n  month = aug,\n  pages = {922--938},\n  publisher = {Association for Computational Linguistics},\n  address = {Online},\n  doi = {10.18653/v1/2021.acl-long.75},\n  url = {https://aclanthology.org/2021.acl-long.75},\n  urldate = {2024-03-19},\n  abstract = {Sequence-to-sequence models excel at handling natural language variation, but have been shown to struggle with out-of-distribution compositional generalization. This has motivated new specialized architectures with stronger compositional biases, but most of these approaches have only been evaluated on synthetically-generated datasets, which are not representative of natural language variation. In this work we ask: can we develop a semantic parsing approach that handles both natural language variation and compositional generalization? To better assess this capability, we propose new train and test splits of non-synthetic datasets. We demonstrate that strong existing approaches do not perform well across a broad set of evaluations. We also propose NQG-T5, a hybrid model that combines a high-precision grammar-based approach with a pre-trained sequence-to-sequence model. It outperforms existing approaches across several compositional generalization challenges on non-synthetic data, while also being competitive with the state-of-the-art on standard evaluations. While still far from solving this problem, our study highlights the importance of diverse evaluations and the open challenge of handling both compositional generalization and natural language variation in semantic parsing.},\n  file = {/Users/shanest/sync/library/Shaw et al/2021/Shaw et al. - 2021 - Compositional Generalization and Natural Language .pdf}\n}\n\n
\n
\n\n\n
\n Sequence-to-sequence models excel at handling natural language variation, but have been shown to struggle with out-of-distribution compositional generalization. This has motivated new specialized architectures with stronger compositional biases, but most of these approaches have only been evaluated on synthetically-generated datasets, which are not representative of natural language variation. In this work we ask: can we develop a semantic parsing approach that handles both natural language variation and compositional generalization? To better assess this capability, we propose new train and test splits of non-synthetic datasets. We demonstrate that strong existing approaches do not perform well across a broad set of evaluations. We also propose NQG-T5, a hybrid model that combines a high-precision grammar-based approach with a pre-trained sequence-to-sequence model. It outperforms existing approaches across several compositional generalization challenges on non-synthetic data, while also being competitive with the state-of-the-art on standard evaluations. While still far from solving this problem, our study highlights the importance of diverse evaluations and the open challenge of handling both compositional generalization and natural language variation in semantic parsing.\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n \n Animal Linguistics: Exploring Referentiality and Compositionality in Bird Calls.\n \n \n \n \n\n\n \n Suzuki, T. N.\n\n\n \n\n\n\n Ecological Research, 36(2): 221–231. 2021.\n \n\n\n\n
\n\n\n\n \n \n \"AnimalPaper\n  \n \n\n \n \n doi\n  \n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n \n \n \n \n \n \n \n \n \n \n\n\n\n
\n 
@article{suzukiAnimalLinguisticsExploring2021,\n  title = {Animal Linguistics: {{Exploring}} Referentiality and Compositionality in Bird Calls},\n  shorttitle = {Animal Linguistics},\n  author = {Suzuki, Toshitaka N.},\n  year = {2021},\n  journal = {Ecological Research},\n  volume = {36},\n  number = {2},\n  pages = {221--231},\n  issn = {1440-1703},\n  doi = {10.1111/1440-1703.12200},\n  url = {https://onlinelibrary.wiley.com/doi/abs/10.1111/1440-1703.12200},\n  urldate = {2024-03-19},\n  abstract = {Establishing the theory of language evolution is an ongoing challenge in science. One profitable approach in this regard is to seek the origins of linguistic capabilities by comparing language with the vocal communication systems of closely related relatives (i.e., the great apes). However, several key capabilities of language appear to be absent in non-human primates, which limits the range of studies, such as direct phylogenetic comparison. A further informative approach lies in identifying convergent features in phylogenetically distant animals and conducting comparative studies. This approach is particularly useful with respect to establishing general rules for the evolution of linguistic capabilities. In this article, I review recent findings on linguistic capabilities in a passerine bird species, the Japanese tit (Parus minor). Field experiments have revealed that Japanese tits produce unique alarm calls when encountering predatory snakes, which serve to enhance the visual attention of call receivers with respect to snake-like objects. Moreover, tits often combine discrete types of meaningful calls into fixed-ordered sequences according to an ordering rule, conveying a compositional message to receivers. These findings indicate that two core capabilities of language, namely, referentiality and compositionality, have independently evolved in the avian lineage. I describe how these linguistic capabilities can be examined under field conditions and discuss how such research may contribute to exploring the origins and evolution of language.},\n  copyright = {{\\copyright} 2021 The Authors. Ecological Research published by John Wiley \\& Sons Australia, Ltd on behalf of The Ecological Society of Japan},\n  langid = {english},\n  keywords = {bird,communication,compositionality,language evolution,referentiality},\n  file = {/Users/shanest/sync/library/Suzuki/2021/Suzuki - 2021 - Animal linguistics Exploring referentiality and c.pdf;/Users/shanest/Zotero/storage/HVEGDXDV/1440-1703.html}\n}\n\n
\n
\n\n\n
\n Establishing the theory of language evolution is an ongoing challenge in science. One profitable approach in this regard is to seek the origins of linguistic capabilities by comparing language with the vocal communication systems of closely related relatives (i.e., the great apes). However, several key capabilities of language appear to be absent in non-human primates, which limits the range of studies, such as direct phylogenetic comparison. A further informative approach lies in identifying convergent features in phylogenetically distant animals and conducting comparative studies. This approach is particularly useful with respect to establishing general rules for the evolution of linguistic capabilities. In this article, I review recent findings on linguistic capabilities in a passerine bird species, the Japanese tit (Parus minor). Field experiments have revealed that Japanese tits produce unique alarm calls when encountering predatory snakes, which serve to enhance the visual attention of call receivers with respect to snake-like objects. Moreover, tits often combine discrete types of meaningful calls into fixed-ordered sequences according to an ordering rule, conveying a compositional message to receivers. These findings indicate that two core capabilities of language, namely, referentiality and compositionality, have independently evolved in the avian lineage. I describe how these linguistic capabilities can be examined under field conditions and discuss how such research may contribute to exploring the origins and evolution of language.\n
\n\n\n
\n\n\n\n\n\n
\n
\n\n
\n
\n  \n 2020\n \n \n (14)\n \n \n
\n
\n \n \n
\n \n\n \n \n \n \n \n \n Good-Enough Compositional Data Augmentation.\n \n \n \n \n\n\n \n Andreas, J.\n\n\n \n\n\n\n In Jurafsky, D.; Chai, J.; Schluter, N.; and Tetreault, J., editor(s), Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 7556–7566, Online, July 2020. Association for Computational Linguistics\n \n\n\n\n
\n\n\n\n \n \n \"Good-EnoughPaper\n  \n \n\n \n \n doi\n  \n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n  \n \n 8 downloads\n \n \n\n \n \n \n \n \n \n \n\n  \n \n \n\n\n\n
\n 
@inproceedings{andreasGoodEnoughCompositionalData2020,\n  title = {Good-{{Enough Compositional Data Augmentation}}},\n  booktitle = {Proceedings of the 58th {{Annual Meeting}} of the {{Association}} for {{Computational Linguistics}}},\n  author = {Andreas, Jacob},\n  editor = {Jurafsky, Dan and Chai, Joyce and Schluter, Natalie and Tetreault, Joel},\n  year = {2020},\n  month = jul,\n  pages = {7556--7566},\n  publisher = {Association for Computational Linguistics},\n  address = {Online},\n  doi = {10.18653/v1/2020.acl-main.676},\n  url = {https://aclanthology.org/2020.acl-main.676},\n  urldate = {2024-03-18},\n  abstract = {We propose a simple data augmentation protocol aimed at providing a compositional inductive bias in conditional and unconditional sequence models. Under this protocol, synthetic training examples are constructed by taking real training examples and replacing (possibly discontinuous) fragments with other fragments that appear in at least one similar environment. The protocol is model-agnostic and useful for a variety of tasks. Applied to neural sequence-to-sequence models, it reduces error rate by as much as 87\\% on diagnostic tasks from the SCAN dataset and 16\\% on a semantic parsing task. Applied to n-gram language models, it reduces perplexity by roughly 1\\% on small corpora in several languages.},\n  file = {/Users/shanest/sync/library/Andreas/2020/Andreas - 2020 - Good-Enough Compositional Data Augmentation.pdf}\n}\n\n
\n
\n\n\n
\n We propose a simple data augmentation protocol aimed at providing a compositional inductive bias in conditional and unconditional sequence models. Under this protocol, synthetic training examples are constructed by taking real training examples and replacing (possibly discontinuous) fragments with other fragments that appear in at least one similar environment. The protocol is model-agnostic and useful for a variety of tasks. Applied to neural sequence-to-sequence models, it reduces error rate by as much as 87% on diagnostic tasks from the SCAN dataset and 16% on a semantic parsing task. Applied to n-gram language models, it reduces perplexity by roughly 1% on small corpora in several languages.\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n \n On the Evolution of Compositional Language.\n \n \n \n \n\n\n \n Barrett, J. A.; Cochran, C.; and Skyrms, B.\n\n\n \n\n\n\n Philosophy of Science, 87(5): 910–920. December 2020.\n \n\n\n\n
\n\n\n\n \n \n \"OnPaper\n  \n \n\n \n \n doi\n  \n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n\n\n\n
\n 
@article{barrettEvolutionCompositionalLanguage2020,\n  title = {On the {{Evolution}} of {{Compositional Language}}},\n  author = {Barrett, Jeffrey A. and Cochran, Calvin and Skyrms, Brian},\n  year = {2020},\n  month = dec,\n  journal = {Philosophy of Science},\n  volume = {87},\n  number = {5},\n  pages = {910--920},\n  issn = {0031-8248, 1539-767X},\n  doi = {10.1086/710367},\n  url = {https://www.cambridge.org/core/journals/philosophy-of-science/article/on-the-evolution-of-compositional-language/E65AF2A9D2DB2B8E3C8B4AA0C7273592},\n  urldate = {2024-04-05},\n  abstract = {We present here a hierarchical model for the evolution of compositional language. The model has the structure of a two-sender/one-receiver Lewis signaling game augmented with executive agents who may learn to influence the behavior of the basic senders and receiver. The model shows how functional agents might coevolve representational roles even as they evolve a reliable compositional language in the context of costly signaling. When successful, the evolved language captures both the compositional structure of properties in the world and the compositional structure of successful actions involving those properties.},\n  langid = {english}\n}\n\n
\n
\n\n\n
\n We present here a hierarchical model for the evolution of compositional language. The model has the structure of a two-sender/one-receiver Lewis signaling game augmented with executive agents who may learn to influence the behavior of the basic senders and receiver. The model shows how functional agents might coevolve representational roles even as they evolve a reliable compositional language in the context of costly signaling. When successful, the evolved language captures both the compositional structure of properties in the world and the compositional structure of successful actions involving those properties.\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n \n Compositionality and Generalization In Emergent Languages.\n \n \n \n \n\n\n \n Chaabouni, R.; Kharitonov, E.; Bouchacourt, D.; Dupoux, E.; and Baroni, M.\n\n\n \n\n\n\n In Jurafsky, D.; Chai, J.; Schluter, N.; and Tetreault, J., editor(s), Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 4427–4442, Online, July 2020. Association for Computational Linguistics\n \n\n\n\n
\n\n\n\n \n \n \"CompositionalityPaper\n  \n \n\n \n \n doi\n  \n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n  \n \n 12 downloads\n \n \n\n \n \n \n \n \n \n \n\n  \n \n \n\n\n\n
\n 
@inproceedings{chaabouniCompositionalityGeneralizationEmergent2020,\n  title = {Compositionality and {{Generalization In Emergent Languages}}},\n  booktitle = {Proceedings of the 58th {{Annual Meeting}} of the {{Association}} for {{Computational Linguistics}}},\n  author = {Chaabouni, Rahma and Kharitonov, Eugene and Bouchacourt, Diane and Dupoux, Emmanuel and Baroni, Marco},\n  editor = {Jurafsky, Dan and Chai, Joyce and Schluter, Natalie and Tetreault, Joel},\n  year = {2020},\n  month = jul,\n  pages = {4427--4442},\n  publisher = {Association for Computational Linguistics},\n  address = {Online},\n  doi = {10.18653/v1/2020.acl-main.407},\n  url = {https://aclanthology.org/2020.acl-main.407},\n  urldate = {2024-03-18},\n  abstract = {Natural language allows us to refer to novel composite concepts by combining expressions denoting their parts according to systematic rules, a property known as compositionality. In this paper, we study whether the language emerging in deep multi-agent simulations possesses a similar ability to refer to novel primitive combinations, and whether it accomplishes this feat by strategies akin to human-language compositionality. Equipped with new ways to measure compositionality in emergent languages inspired by disentanglement in representation learning, we establish three main results: First, given sufficiently large input spaces, the emergent language will naturally develop the ability to refer to novel composite concepts. Second, there is no correlation between the degree of compositionality of an emergent language and its ability to generalize. Third, while compositionality is not necessary for generalization, it provides an advantage in terms of language transmission: The more compositional a language is, the more easily it will be picked up by new learners, even when the latter differ in architecture from the original agents. We conclude that compositionality does not arise from simple generalization pressure, but if an emergent language does chance upon it, it will be more likely to survive and thrive.},\n  file = {/Users/shanest/sync/library/Chaabouni et al/2020/Chaabouni et al. - 2020 - Compositionality and Generalization In Emergent La3.pdf}\n}\n\n
\n
\n\n\n
\n Natural language allows us to refer to novel composite concepts by combining expressions denoting their parts according to systematic rules, a property known as compositionality. In this paper, we study whether the language emerging in deep multi-agent simulations possesses a similar ability to refer to novel primitive combinations, and whether it accomplishes this feat by strategies akin to human-language compositionality. Equipped with new ways to measure compositionality in emergent languages inspired by disentanglement in representation learning, we establish three main results: First, given sufficiently large input spaces, the emergent language will naturally develop the ability to refer to novel composite concepts. Second, there is no correlation between the degree of compositionality of an emergent language and its ability to generalize. Third, while compositionality is not necessary for generalization, it provides an advantage in terms of language transmission: The more compositional a language is, the more easily it will be picked up by new learners, even when the latter differ in architecture from the original agents. We conclude that compositionality does not arise from simple generalization pressure, but if an emergent language does chance upon it, it will be more likely to survive and thrive.\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n \n Compositional Generalization via Neural-Symbolic Stack Machines.\n \n \n \n \n\n\n \n Chen, X.; Liang, C.; Yu, A. W.; Song, D.; and Zhou, D.\n\n\n \n\n\n\n In Advances in Neural Information Processing Systems, volume 33, pages 1690–1701, 2020. Curran Associates, Inc.\n \n\n\n\n
\n\n\n\n \n \n \"CompositionalPaper\n  \n \n\n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n\n\n\n
\n 
@inproceedings{chenCompositionalGeneralizationNeuralSymbolic2020,\n  title = {Compositional {{Generalization}} via {{Neural-Symbolic Stack Machines}}},\n  booktitle = {Advances in {{Neural Information Processing Systems}}},\n  author = {Chen, Xinyun and Liang, Chen and Yu, Adams Wei and Song, Dawn and Zhou, Denny},\n  year = {2020},\n  volume = {33},\n  pages = {1690--1701},\n  publisher = {Curran Associates, Inc.},\n  url = {https://proceedings.neurips.cc/paper/2020/hash/12b1e42dc0746f22cf361267de07073f-Abstract.html},\n  urldate = {2024-03-19},\n  abstract = {Despite achieving tremendous success, existing deep learning models have exposed limitations in compositional generalization, the capability to learn compositional rules and apply them to unseen cases in a systematic manner. To tackle this issue, we propose the Neural-Symbolic Stack Machine (NeSS). It contains a neural network to generate traces, which are then executed by a symbolic stack machine enhanced with sequence manipulation operations. NeSS combines the expressive power of neural sequence models with the recursion supported by the symbolic stack machine. Without training supervision on execution traces, NeSS achieves 100\\% generalization performance in four domains: the SCAN benchmark of language-driven navigation tasks, the task of few-shot learning of compositional instructions, the compositional machine translation benchmark, and context-free grammar parsing tasks.},\n  file = {/Users/shanest/sync/library/Chen et al/2020/Chen et al. - 2020 - Compositional Generalization via Neural-Symbolic S.pdf}\n}\n\n
\n
\n\n\n
\n Despite achieving tremendous success, existing deep learning models have exposed limitations in compositional generalization, the capability to learn compositional rules and apply them to unseen cases in a systematic manner. To tackle this issue, we propose the Neural-Symbolic Stack Machine (NeSS). It contains a neural network to generate traces, which are then executed by a symbolic stack machine enhanced with sequence manipulation operations. NeSS combines the expressive power of neural sequence models with the recursion supported by the symbolic stack machine. Without training supervision on execution traces, NeSS achieves 100% generalization performance in four domains: the SCAN benchmark of language-driven navigation tasks, the task of few-shot learning of compositional instructions, the compositional machine translation benchmark, and context-free grammar parsing tasks.\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n \n Permutation Equivariant Models for Compositional Generalization in Language.\n \n \n \n \n\n\n \n Gordon, J.; Lopez-Paz, D.; Baroni, M.; and Bouchacourt, D.\n\n\n \n\n\n\n In International Conference on Learning Representations, 2020. \n \n\n\n\n
\n\n\n\n \n \n \"PermutationPaper\n  \n \n\n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n\n\n\n
\n 
@inproceedings{gordonPermutationEquivariantModels2020,\n  title = {Permutation {{Equivariant Models}} for {{Compositional Generalization}} in {{Language}}},\n  booktitle = {International {{Conference}} on {{Learning Representations}}},\n  author = {Gordon, Jonathan and {Lopez-Paz}, David and Baroni, Marco and Bouchacourt, Diane},\n  year = {2020},\n  url = {https://openreview.net/forum?id=SylVNerFvr},\n  urldate = {2024-03-19},\n  abstract = {Humans understand novel sentences by composing meanings and roles of core language components. In contrast, neural network models for natural language modeling fail when such compositional generalization is required. The main contribution of this paper is to hypothesize that language compositionality is a form of group-equivariance. Based on this hypothesis, we propose a set of tools for constructing equivariant sequence-to-sequence models. Throughout a variety of experiments on the SCAN tasks, we analyze the behavior of existing models under the lens of equivariance, and demonstrate that our equivariant architecture is able to achieve the type compositional generalization required in human language understanding.},\n  langid = {english},\n  file = {/Users/shanest/sync/library/Gordon et al/2019/Gordon et al. - 2019 - Permutation Equivariant Models for Compositional G.pdf}\n}\n\n
\n
\n\n\n
\n Humans understand novel sentences by composing meanings and roles of core language components. In contrast, neural network models for natural language modeling fail when such compositional generalization is required. The main contribution of this paper is to hypothesize that language compositionality is a form of group-equivariance. Based on this hypothesis, we propose a set of tools for constructing equivariant sequence-to-sequence models. Throughout a variety of experiments on the SCAN tasks, we analyze the behavior of existing models under the lens of equivariance, and demonstrate that our equivariant architecture is able to achieve the type compositional generalization required in human language understanding.\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n \n Compositionality Decomposed: How Do Neural Networks Generalise?.\n \n \n \n \n\n\n \n Hupkes, D.; Dankers, V.; Mul, M.; and Bruni, E.\n\n\n \n\n\n\n Journal of Artificial Intelligence Research, 67: 757–795. April 2020.\n \n\n\n\n
\n\n\n\n \n \n \"CompositionalityPaper\n  \n \n\n \n \n doi\n  \n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n  \n \n 11 downloads\n \n \n\n \n \n \n \n \n \n \n\n  \n \n \n \n \n \n \n \n \n \n \n\n\n\n
\n 
@article{hupkesCompositionalityDecomposedHow2020,\n  title = {Compositionality {{Decomposed}}: {{How}} Do {{Neural Networks Generalise}}?},\n  shorttitle = {Compositionality {{Decomposed}}},\n  author = {Hupkes, Dieuwke and Dankers, Verna and Mul, Mathijs and Bruni, Elia},\n  year = {2020},\n  month = apr,\n  journal = {Journal of Artificial Intelligence Research},\n  volume = {67},\n  pages = {757--795},\n  issn = {1076-9757},\n  doi = {10.1613/jair.1.11674},\n  url = {https://www.jair.org/index.php/jair/article/view/11674},\n  urldate = {2024-03-18},\n  abstract = {Despite a multitude of empirical studies, little consensus exists on whether neural networks are able to generalise compositionally, a controversy that, in part, stems from a lack of agreement about what it means for a neural model to be compositional. As a response to this controversy, we present a set of tests that provide a bridge between, on the one hand, the vast amount of linguistic and philosophical theory about compositionality of language and, on the other, the successful neural models of language. We collect different interpretations of compositionality and translate them into five theoretically grounded tests for models that are formulated on a task-independent level. In particular, we provide tests to investigate (i) if models systematically recombine known parts and rules (ii) if models can extend their predictions beyond the length they have seen in the training data (iii) if models' composition operations are local or global (iv) if models' predictions are robust to synonym substitutions and (v) if models favour rules or exceptions during training. To demonstrate the usefulness of this evaluation paradigm, we instantiate these five tests on a highly compositional data set which we dub PCFG SET and apply the resulting tests to three popular sequence-to-sequence models: a recurrent, a convolution-based and a transformer model. We provide an in-depth analysis of the results, which uncover the strengths and weaknesses of these three architectures and point to potential areas of improvement.},\n  copyright = {Copyright (c)},\n  langid = {english},\n  keywords = {machine learning,natural language,neural networks,rule learning},\n  file = {/Users/shanest/sync/library/Hupkes et al/2020/Hupkes et al. - 2020 - Compositionality Decomposed How do Neural Network.pdf}\n}\n\n
\n
\n\n\n
\n Despite a multitude of empirical studies, little consensus exists on whether neural networks are able to generalise compositionally, a controversy that, in part, stems from a lack of agreement about what it means for a neural model to be compositional. As a response to this controversy, we present a set of tests that provide a bridge between, on the one hand, the vast amount of linguistic and philosophical theory about compositionality of language and, on the other, the successful neural models of language. We collect different interpretations of compositionality and translate them into five theoretically grounded tests for models that are formulated on a task-independent level. In particular, we provide tests to investigate (i) if models systematically recombine known parts and rules (ii) if models can extend their predictions beyond the length they have seen in the training data (iii) if models' composition operations are local or global (iv) if models' predictions are robust to synonym substitutions and (v) if models favour rules or exceptions during training. To demonstrate the usefulness of this evaluation paradigm, we instantiate these five tests on a highly compositional data set which we dub PCFG SET and apply the resulting tests to three popular sequence-to-sequence models: a recurrent, a convolution-based and a transformer model. We provide an in-depth analysis of the results, which uncover the strengths and weaknesses of these three architectures and point to potential areas of improvement.\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n \n Measuring Compositional Generalization: A Comprehensive Method on Realistic Data.\n \n \n \n \n\n\n \n Keysers, D.; Schärli, N.; Scales, N.; Buisman, H.; Furrer, D.; Kashubin, S.; Momchev, N.; Sinopalnikov, D.; Stafiniak, L.; Tihon, T.; Tsarkov, D.; Wang, X.; van Zee, M.; and Bousquet, O.\n\n\n \n\n\n\n In International Conference on Learning Representations, 2020. \n \n\n\n\n
\n\n\n\n \n \n \"MeasuringPaper\n  \n \n\n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n  \n \n 13 downloads\n \n \n\n \n \n \n \n \n \n \n\n  \n \n \n\n\n\n
\n 
@inproceedings{keysersMeasuringCompositionalGeneralization2020,\n  title = {Measuring {{Compositional Generalization}}: {{A Comprehensive Method}} on {{Realistic Data}}},\n  shorttitle = {Measuring {{Compositional Generalization}}},\n  booktitle = {International {{Conference}} on {{Learning Representations}}},\n  author = {Keysers, Daniel and Sch{\\"a}rli, Nathanael and Scales, Nathan and Buisman, Hylke and Furrer, Daniel and Kashubin, Sergii and Momchev, Nikola and Sinopalnikov, Danila and Stafiniak, Lukasz and Tihon, Tibor and Tsarkov, Dmitry and Wang, Xiao and van Zee, Marc and Bousquet, Olivier},\n  year = {2020},\n  url = {https://openreview.net/forum?id=SygcCnNKwr},\n  urldate = {2024-03-18},\n  abstract = {State-of-the-art machine learning methods exhibit limited compositional generalization. At the same time, there is a lack of realistic benchmarks that comprehensively measure this ability, which makes it challenging to find and evaluate improvements. We introduce a novel method to systematically construct such benchmarks by maximizing compound divergence while guaranteeing a small atom divergence between train and test sets, and we quantitatively compare this method to other approaches for creating compositional generalization benchmarks. We present a large and realistic natural language question answering dataset that is constructed according to this method, and we use it to analyze the compositional generalization ability of three machine learning architectures. We find that they fail to generalize compositionally and that there is a surprisingly strong negative correlation between compound divergence and accuracy. We also demonstrate how our method can be used to create new compositionality benchmarks on top of the existing SCAN dataset, which confirms these findings.},\n  langid = {english},\n  file = {/Users/shanest/sync/library/Keysers et al/2019/Keysers et al. - 2020 - Measuring Compositional Generalization A Comprehe.pdf}\n}\n\n
\n
\n\n\n
\n State-of-the-art machine learning methods exhibit limited compositional generalization. At the same time, there is a lack of realistic benchmarks that comprehensively measure this ability, which makes it challenging to find and evaluate improvements. We introduce a novel method to systematically construct such benchmarks by maximizing compound divergence while guaranteeing a small atom divergence between train and test sets, and we quantitatively compare this method to other approaches for creating compositional generalization benchmarks. We present a large and realistic natural language question answering dataset that is constructed according to this method, and we use it to analyze the compositional generalization ability of three machine learning architectures. We find that they fail to generalize compositionally and that there is a surprisingly strong negative correlation between compound divergence and accuracy. We also demonstrate how our method can be used to create new compositionality benchmarks on top of the existing SCAN dataset, which confirms these findings.\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n \n What They Do When in Doubt: A Study of Inductive Biases in Seq2seq Learners.\n \n \n \n \n\n\n \n Kharitonov, E.; and Chaabouni, R.\n\n\n \n\n\n\n In International Conference on Learning Representations, October 2020. \n \n\n\n\n
\n\n\n\n \n \n \"WhatPaper\n  \n \n\n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n  \n \n 4 downloads\n \n \n\n \n \n \n \n \n \n \n\n  \n \n \n\n\n\n
\n 
@inproceedings{kharitonovWhatTheyWhen2020,\n  title = {What They Do When in Doubt: A Study of Inductive Biases in Seq2seq Learners},\n  shorttitle = {What They Do When in Doubt},\n  booktitle = {International {{Conference}} on {{Learning Representations}}},\n  author = {Kharitonov, Eugene and Chaabouni, Rahma},\n  year = {2020},\n  month = oct,\n  url = {https://openreview.net/forum?id=YmA86Zo-P_t},\n  urldate = {2024-03-19},\n  abstract = {Sequence-to-sequence (seq2seq) learners are widely used, but we still have only limited knowledge about what inductive biases shape the way they generalize. We address that by investigating how popular seq2seq learners generalize in tasks that have high ambiguity in the training data. We use four new tasks to study learners' preferences for memorization, arithmetic, hierarchical, and compositional reasoning. Further, we connect to Solomonoff's theory of induction and propose to use description length as a principled and sensitive measure of inductive biases. In our experimental study, we find that LSTM-based learners can learn to perform counting, addition, and multiplication by a constant from a single training example. Furthermore, Transformer and LSTM-based learners show a bias toward the hierarchical induction over the linear one, while CNN-based learners prefer the opposite. The latter also show a bias toward a compositional generalization over memorization. Finally, across all our experiments, description length proved to be a sensitive measure of inductive biases.},\n  langid = {english},\n  file = {/Users/shanest/Zotero/storage/GCYIGIP8/Kharitonov and Chaabouni - 2020 - What they do when in doubt a study of inductive b.pdf}\n}\n\n
\n
\n\n\n
\n Sequence-to-sequence (seq2seq) learners are widely used, but we still have only limited knowledge about what inductive biases shape the way they generalize. We address that by investigating how popular seq2seq learners generalize in tasks that have high ambiguity in the training data. We use four new tasks to study learners' preferences for memorization, arithmetic, hierarchical, and compositional reasoning. Further, we connect to Solomonoff's theory of induction and propose to use description length as a principled and sensitive measure of inductive biases. In our experimental study, we find that LSTM-based learners can learn to perform counting, addition, and multiplication by a constant from a single training example. Furthermore, Transformer and LSTM-based learners show a bias toward the hierarchical induction over the linear one, while CNN-based learners prefer the opposite. The latter also show a bias toward a compositional generalization over memorization. Finally, across all our experiments, description length proved to be a sensitive measure of inductive biases.\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n \n COGS: A Compositional Generalization Challenge Based on Semantic Interpretation.\n \n \n \n \n\n\n \n Kim, N.; and Linzen, T.\n\n\n \n\n\n\n In Webber, B.; Cohn, T.; He, Y.; and Liu, Y., editor(s), Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 9087–9105, Online, November 2020. Association for Computational Linguistics\n \n\n\n\n
\n\n\n\n \n \n \"COGS:Paper\n  \n \n\n \n \n doi\n  \n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n  \n \n 11 downloads\n \n \n\n \n \n \n \n \n \n \n\n  \n \n \n\n\n\n
\n 
@inproceedings{kimCOGSCompositionalGeneralization2020,\n  title = {{{COGS}}: {{A Compositional Generalization Challenge Based}} on {{Semantic Interpretation}}},\n  shorttitle = {{{COGS}}},\n  booktitle = {Proceedings of the 2020 {{Conference}} on {{Empirical Methods}} in {{Natural Language Processing}} ({{EMNLP}})},\n  author = {Kim, Najoung and Linzen, Tal},\n  editor = {Webber, Bonnie and Cohn, Trevor and He, Yulan and Liu, Yang},\n  year = {2020},\n  month = nov,\n  pages = {9087--9105},\n  publisher = {Association for Computational Linguistics},\n  address = {Online},\n  doi = {10.18653/v1/2020.emnlp-main.731},\n  url = {https://aclanthology.org/2020.emnlp-main.731},\n  urldate = {2024-03-18},\n  abstract = {Natural language is characterized by compositionality: the meaning of a complex expression is constructed from the meanings of its constituent parts. To facilitate the evaluation of the compositional abilities of language processing architectures, we introduce COGS, a semantic parsing dataset based on a fragment of English. The evaluation portion of COGS contains multiple systematic gaps that can only be addressed by compositional generalization; these include new combinations of familiar syntactic structures, or new combinations of familiar words and familiar structures. In experiments with Transformers and LSTMs, we found that in-distribution accuracy on the COGS test set was near-perfect (96--99\\%), but generalization accuracy was substantially lower (16--35\\%) and showed high sensitivity to random seed (+-6--8\\%). These findings indicate that contemporary standard NLP models are limited in their compositional generalization capacity, and position COGS as a good way to measure progress.},\n  file = {/Users/shanest/sync/library/Kim_Linzen/2020/Kim and Linzen - 2020 - COGS A Compositional Generalization Challenge Bas.pdf}\n}\n\n
\n
\n\n\n
\n Natural language is characterized by compositionality: the meaning of a complex expression is constructed from the meanings of its constituent parts. To facilitate the evaluation of the compositional abilities of language processing architectures, we introduce COGS, a semantic parsing dataset based on a fragment of English. The evaluation portion of COGS contains multiple systematic gaps that can only be addressed by compositional generalization; these include new combinations of familiar syntactic structures, or new combinations of familiar words and familiar structures. In experiments with Transformers and LSTMs, we found that in-distribution accuracy on the COGS test set was near-perfect (96–99%), but generalization accuracy was substantially lower (16–35%) and showed high sensitivity to random seed (+-6–8%). These findings indicate that contemporary standard NLP models are limited in their compositional generalization capacity, and position COGS as a good way to measure progress.\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n Compositional Generalization by Learning Analytical Expressions.\n \n \n \n\n\n \n Liu, Q.; An, S.; Lou, J.; Chen, B.; Lin, Z.; Gao, Y.; Zhou, B.; Zheng, N.; and Zhang, D.\n\n\n \n\n\n\n In Proceedings of the 34th International Conference on Neural Information Processing Systems, of NIPS'20, pages 11416–11427, Red Hook, NY, USA, December 2020. Curran Associates Inc.\n \n\n\n\n
\n\n\n\n \n\n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n\n\n\n
\n 
@inproceedings{liuCompositionalGeneralizationLearning2020,\n  title = {Compositional Generalization by Learning Analytical Expressions},\n  booktitle = {Proceedings of the 34th {{International Conference}} on {{Neural Information Processing Systems}}},\n  author = {Liu, Qian and An, Shengnan and Lou, Jian-Guang and Chen, Bei and Lin, Zeqi and Gao, Yan and Zhou, Bin and Zheng, Nanning and Zhang, Dongmei},\n  year = {2020},\n  month = dec,\n  series = {{{NIPS}}'20},\n  pages = {11416--11427},\n  publisher = {Curran Associates Inc.},\n  address = {Red Hook, NY, USA},\n  urldate = {2024-03-19},\n  abstract = {Compositional generalization is a basic and essential intellective capability of human beings, which allows us to recombine known parts readily. However, existing neural network based models have been proven to be extremely deficient in such a capability. Inspired by work in cognition which argues compositionality can be captured by variable slots with symbolic functions, we present a refreshing view that connects a memory-augmented neural model with analytical expressions, to achieve compositional generalization. Our model consists of two cooperative neural modules, Composer and Solver, fitting well with the cognitive argument while being able to be trained in an end-to-end manner via a hierarchical reinforcement learning algorithm. Experiments on the well-known benchmark SCAN demonstrate that our model seizes a great ability of compositional generalization, solving all challenges addressed by previous works with 100\\% accuracies.},\n  isbn = {978-1-71382-954-6},\n  file = {/Users/shanest/sync/library/Liu et al/2020/Liu et al. - 2020 - Compositional generalization by learning analytica.pdf}\n}\n\n
\n
\n\n\n
\n Compositional generalization is a basic and essential intellective capability of human beings, which allows us to recombine known parts readily. However, existing neural network based models have been proven to be extremely deficient in such a capability. Inspired by work in cognition which argues compositionality can be captured by variable slots with symbolic functions, we present a refreshing view that connects a memory-augmented neural model with analytical expressions, to achieve compositional generalization. Our model consists of two cooperative neural modules, Composer and Solver, fitting well with the cognitive argument while being able to be trained in an end-to-end manner via a hierarchical reinforcement learning algorithm. Experiments on the well-known benchmark SCAN demonstrate that our model seizes a great ability of compositional generalization, solving all challenges addressed by previous works with 100% accuracies.\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n \n Learning Compositional Rules via Neural Program Synthesis.\n \n \n \n \n\n\n \n Nye, M.; Solar-Lezama, A.; Tenenbaum, J.; and Lake, B. M\n\n\n \n\n\n\n In Advances in Neural Information Processing Systems, volume 33, pages 10832–10842, 2020. Curran Associates, Inc.\n \n\n\n\n
\n\n\n\n \n \n \"LearningPaper\n  \n \n\n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n\n\n\n
\n 
@inproceedings{nyeLearningCompositionalRules2020,\n  title = {Learning {{Compositional Rules}} via {{Neural Program Synthesis}}},\n  booktitle = {Advances in {{Neural Information Processing Systems}}},\n  author = {Nye, Maxwell and {Solar-Lezama}, Armando and Tenenbaum, Josh and Lake, Brenden M},\n  year = {2020},\n  volume = {33},\n  pages = {10832--10842},\n  publisher = {Curran Associates, Inc.},\n  url = {https://papers.nips.cc/paper/2020/hash/7a685d9edd95508471a9d3d6fcace432-Abstract.html},\n  urldate = {2024-03-19},\n  abstract = {Many aspects of human reasoning, including language, require learning rules from very little data. Humans can do this, often learning systematic rules from very few examples, and combining these rules to form compositional rule-based systems. Current neural architectures, on the other hand, often fail to generalize in a compositional manner, especially when evaluated in ways that vary systematically from training. In this work, we present a neuro-symbolic model which learns entire rule systems from a small set of examples. Instead of directly predicting outputs from inputs, we train our model to induce the explicit system of rules governing a set of previously seen examples, drawing upon techniques from the neural program synthesis literature. Our rule-synthesis approach outperforms neural meta-learning techniques in three domains: an artificial instruction-learning domain used to evaluate human learning, the SCAN challenge datasets, and learning rule-based translations of number words into integers for a wide range of human languages.},\n  file = {/Users/shanest/sync/library/Nye et al/2020/Nye et al. - 2020 - Learning Compositional Rules via Neural Program Sy.pdf}\n}\n\n
\n
\n\n\n
\n Many aspects of human reasoning, including language, require learning rules from very little data. Humans can do this, often learning systematic rules from very few examples, and combining these rules to form compositional rule-based systems. Current neural architectures, on the other hand, often fail to generalize in a compositional manner, especially when evaluated in ways that vary systematically from training. In this work, we present a neuro-symbolic model which learns entire rule systems from a small set of examples. Instead of directly predicting outputs from inputs, we train our model to induce the explicit system of rules governing a set of previously seen examples, drawing upon techniques from the neural program synthesis literature. Our rule-synthesis approach outperforms neural meta-learning techniques in three domains: an artificial instruction-learning domain used to evaluate human learning, the SCAN challenge datasets, and learning rule-based translations of number words into integers for a wide range of human languages.\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n The Role of Social Network Structure in the Emergence of Linguistic Structure.\n \n \n \n\n\n \n Raviv, L.; Meyer, A.; and Lev-Ari, S.\n\n\n \n\n\n\n Cognitive Science, 44(8). 2020.\n \n\n\n\n
\n\n\n\n \n\n \n \n doi\n  \n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n\n\n\n
\n 
@article{ravivRoleSocialNetwork2020,\n  title = {The {{Role}} of {{Social Network Structure}} in the {{Emergence}} of {{Linguistic Structure}}},\n  author = {Raviv, Limor and Meyer, Antje and {Lev-Ari}, Shiri},\n  year = {2020},\n  journal = {Cognitive Science},\n  volume = {44},\n  number = {8},\n  issn = {15516709},\n  doi = {10.1111/cogs.12876},\n  abstract = {Social network structure has been argued to shape the structure of languages, as well as affect the spread of innovations and the formation of conventions in the community. Specifically, theoretical and computational models of language change predict that sparsely connected communities develop more systematic languages, while tightly knit communities can maintain high levels of linguistic complexity and variability. However, the role of social network structure in the cultural evolution of languages has never been tested experimentally. Here, we present results from a behavioral group communication study, in which we examined the formation of new languages created in the lab by micro-societies that varied in their network structure. We contrasted three types of social networks: fully connected, small-world, and scale-free. We examined the artificial languages created by these different networks with respect to their linguistic structure, communicative success, stability, and convergence. Results did not reveal any effect of network structure for any measure, with all languages becoming similarly more systematic, more accurate, more stable, and more shared over time. At the same time, small-world networks showed the greatest variation in their convergence, stabilization, and emerging structure patterns, indicating that network structure can influence the community's susceptibility to random linguistic changes (i.e., drift).},\n  pmid = {32808326},\n  keywords = {Grammatical structure,Input variability,Language evolution,Linguistic diversity,Network structure,Social structure},\n  file = {/Users/shanest/sync/library/Raviv et al/2020/Raviv et al. - 2020 - The Role of Social Network Structure in the Emerge.pdf}\n}\n\n
\n
\n\n\n
\n Social network structure has been argued to shape the structure of languages, as well as affect the spread of innovations and the formation of conventions in the community. Specifically, theoretical and computational models of language change predict that sparsely connected communities develop more systematic languages, while tightly knit communities can maintain high levels of linguistic complexity and variability. However, the role of social network structure in the cultural evolution of languages has never been tested experimentally. Here, we present results from a behavioral group communication study, in which we examined the formation of new languages created in the lab by micro-societies that varied in their network structure. We contrasted three types of social networks: fully connected, small-world, and scale-free. We examined the artificial languages created by these different networks with respect to their linguistic structure, communicative success, stability, and convergence. Results did not reveal any effect of network structure for any measure, with all languages becoming similarly more systematic, more accurate, more stable, and more shared over time. At the same time, small-world networks showed the greatest variation in their convergence, stabilization, and emerging structure patterns, indicating that network structure can influence the community's susceptibility to random linguistic changes (i.e., drift).\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n \n A Benchmark for Systematic Generalization in Grounded Language Understanding.\n \n \n \n \n\n\n \n Ruis, L.; Andreas, J.; Baroni, M.; Bouchacourt, D.; and Lake, B. M\n\n\n \n\n\n\n In Advances in Neural Information Processing Systems, volume 33, pages 19861–19872, 2020. Curran Associates, Inc.\n \n\n\n\n
\n\n\n\n \n \n \"APaper\n  \n \n\n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n  \n \n 8 downloads\n \n \n\n \n \n \n \n \n \n \n\n  \n \n \n\n\n\n
\n 
@inproceedings{ruisBenchmarkSystematicGeneralization2020,\n  title = {A {{Benchmark}} for {{Systematic Generalization}} in {{Grounded Language Understanding}}},\n  booktitle = {Advances in {{Neural Information Processing Systems}}},\n  author = {Ruis, Laura and Andreas, Jacob and Baroni, Marco and Bouchacourt, Diane and Lake, Brenden M},\n  year = {2020},\n  volume = {33},\n  pages = {19861--19872},\n  publisher = {Curran Associates, Inc.},\n  url = {https://proceedings.neurips.cc/paper/2020/hash/e5a90182cc81e12ab5e72d66e0b46fe3-Abstract.html},\n  urldate = {2024-03-18},\n  abstract = {Humans easily interpret expressions that describe unfamiliar situations composed from familiar parts ("greet the pink brontosaurus by the ferris wheel"). Modern neural networks, by contrast, struggle to interpret novel compositions. In this paper, we introduce a new benchmark, gSCAN, for evaluating compositional generalization in situated language understanding. Going beyond a related benchmark that focused on syntactic aspects of generalization, gSCAN defines a language grounded in the states of a grid world, facilitating novel evaluations of acquiring linguistically motivated rules. For example, agents must understand how adjectives such as 'small' are interpreted relative to the current world state or how adverbs such as 'cautiously' combine with new verbs. We test a strong multi-modal baseline model and a state-of-the-art compositional method finding that, in most cases, they fail dramatically when generalization requires systematic compositional rules.},\n  file = {/Users/shanest/sync/library/Ruis et al/2020/Ruis et al. - 2020 - A Benchmark for Systematic Generalization in Groun.pdf}\n}\n\n
\n
\n\n\n
\n Humans easily interpret expressions that describe unfamiliar situations composed from familiar parts (\"greet the pink brontosaurus by the ferris wheel\"). Modern neural networks, by contrast, struggle to interpret novel compositions. In this paper, we introduce a new benchmark, gSCAN, for evaluating compositional generalization in situated language understanding. Going beyond a related benchmark that focused on syntactic aspects of generalization, gSCAN defines a language grounded in the states of a grid world, facilitating novel evaluations of acquiring linguistically motivated rules. For example, agents must understand how adjectives such as 'small' are interpreted relative to the current world state or how adverbs such as 'cautiously' combine with new verbs. We test a strong multi-modal baseline model and a state-of-the-art compositional method finding that, in most cases, they fail dramatically when generalization requires systematic compositional rules.\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n \n Toward the Emergence of Nontrivial Compositionality.\n \n \n \n \n\n\n \n Steinert-Threlkeld, S.\n\n\n \n\n\n\n Philosophy of Science, 87(5): 897–909. 2020.\n \n\n\n\n
\n\n\n\n \n \n \"TowardPaper\n  \n \n\n \n \n doi\n  \n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n  \n \n 1 download\n \n \n\n \n \n \n \n \n \n \n\n  \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n\n\n\n
\n 
@article{steinert-threlkeldEmergenceNontrivialCompositionality2020,\n  title = {Toward the {{Emergence}} of {{Nontrivial Compositionality}}},\n  author = {{Steinert-Threlkeld}, Shane},\n  year = {2020},\n  journal = {Philosophy of Science},\n  volume = {87},\n  number = {5},\n  pages = {897--909},\n  publisher = {The University of Chicago Press},\n  issn = {0031-8248},\n  doi = {10.1086/710628},\n  url = {https://www.journals.uchicago.edu/doi/10.1086/710628},\n  urldate = {2022-03-02},\n  abstract = {All natural languages exhibit a distinction between content words (nouns, verbs, etc.) and function words (determiners, auxiliaries, tenses, etc.). Yet surprisingly little has been said about the emergence of this universal architectural feature of human language. This article argues that the existence of this distinction requires the presence of nontrivial compositionality and identifies assumptions that have previously been made in the literature that provably guarantee only trivial composition. It then presents a signaling game with variable contexts and shows how the distinction can emerge via reinforcement learning.},\n  keywords = {Artificial Intelligence,Cognitive Science,evolution,function,games,language,learning,networks,neural,of,reinforcement,signaling,words},\n  file = {/Users/shanest/sync/library/Steinert-Threlkeld/2020/Steinert-Threlkeld - 2020 - Toward the Emergence of Nontrivial Compositionalit.pdf}\n}\n\n
\n
\n\n\n
\n All natural languages exhibit a distinction between content words (nouns, verbs, etc.) and function words (determiners, auxiliaries, tenses, etc.). Yet surprisingly little has been said about the emergence of this universal architectural feature of human language. This article argues that the existence of this distinction requires the presence of nontrivial compositionality and identifies assumptions that have previously been made in the literature that provably guarantee only trivial composition. It then presents a signaling game with variable contexts and shows how the distinction can emerge via reinforcement learning.\n
\n\n\n
\n\n\n\n\n\n
\n
\n\n
\n
\n  \n 2019\n \n \n (6)\n \n \n
\n
\n \n \n
\n \n\n \n \n \n \n \n \n Measuring Compositionality in Representation Learning.\n \n \n \n \n\n\n \n Andreas, J.\n\n\n \n\n\n\n In International Conference of Learning Representations, 2019. \n \n\n\n\n
\n\n\n\n \n \n \"MeasuringPaper\n  \n \n\n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n  \n \n 14 downloads\n \n \n\n \n \n \n \n \n \n \n\n  \n \n \n \n \n \n \n\n\n\n
\n 
@inproceedings{andreasMeasuringCompositionalityRepresentation2019,\n  title = {Measuring {{Compositionality}} in {{Representation Learning}}},\n  booktitle = {International {{Conference}} of {{Learning Representations}}},\n  author = {Andreas, Jacob},\n  year = {2019},\n  eprint = {1902.07181},\n  url = {https://openreview.net/forum?id=HJz05o0qK7},\n  abstract = {Many machine learning algorithms represent input data with vector embeddings or discrete codes. When inputs exhibit compositional structure (e.g. objects built from parts or procedures from subroutines), it is natural to ask whether this compositional structure is reflected in the the inputs' learned representations. While the assessment of compositionality in languages has received significant attention in linguistics and adjacent fields, the machine learning literature lacks general-purpose tools for producing graded measurements of compositional structure in more general (e.g. vector-valued) representation spaces. We describe a procedure for evaluating compositionality by measuring how well the true representation-producing model can be approximated by a model that explicitly composes a collection of inferred representational primitives. We use the procedure to provide formal and empirical characterizations of compositional structure in a variety of settings, exploring the relationship between compositionality and learning dynamics, human judgments, representational similarity, and generalization.},\n  archiveprefix = {arxiv},\n  keywords = {method: tree reconstruction error,phenomenon: compositionality},\n  file = {/Users/shanest/sync/library/Andreas/2019/Andreas - 2019 - Measuring Compositionality in Representation Learn.pdf}\n}\n\n
\n
\n\n\n
\n Many machine learning algorithms represent input data with vector embeddings or discrete codes. When inputs exhibit compositional structure (e.g. objects built from parts or procedures from subroutines), it is natural to ask whether this compositional structure is reflected in the the inputs' learned representations. While the assessment of compositionality in languages has received significant attention in linguistics and adjacent fields, the machine learning literature lacks general-purpose tools for producing graded measurements of compositional structure in more general (e.g. vector-valued) representation spaces. We describe a procedure for evaluating compositionality by measuring how well the true representation-producing model can be approximated by a model that explicitly composes a collection of inferred representational primitives. We use the procedure to provide formal and empirical characterizations of compositional structure in a variety of settings, exploring the relationship between compositionality and learning dynamics, human judgments, representational similarity, and generalization.\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n \n Human Few-Shot Learning of Compositional Instructions.\n \n \n \n \n\n\n \n Lake, B. M.; Linzen, T.; and Baroni, M.\n\n\n \n\n\n\n In Proceedings of the 41st Annual Conference of the Cognitive Science Society, May 2019. \n \n\n\n\n
\n\n\n\n \n \n \"HumanPaper\n  \n \n\n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n  \n \n 1 download\n \n \n\n \n \n \n \n \n \n \n\n  \n \n \n \n \n\n\n\n
\n 
@inproceedings{lakeHumanFewshotLearning2019,\n  title = {Human Few-Shot Learning of Compositional Instructions},\n  booktitle = {Proceedings of the 41st {{Annual Conference}} of the {{Cognitive Science Society}}},\n  author = {Lake, Brenden M. and Linzen, Tal and Baroni, Marco},\n  year = {2019},\n  month = may,\n  eprint = {1901.04587},\n  primaryclass = {cs},\n  url = {http://arxiv.org/abs/1901.04587},\n  urldate = {2024-03-19},\n  abstract = {People learn in fast and flexible ways that have not been emulated by machines. Once a person learns a new verb "dax," he or she can effortlessly understand how to "dax twice," "walk and dax," or "dax vigorously." There have been striking recent improvements in machine learning for natural language processing, yet the best algorithms require vast amounts of experience and struggle to generalize new concepts in compositional ways. To better understand these distinctively human abilities, we study the compositional skills of people through language-like instruction learning tasks. Our results show that people can learn and use novel functional concepts from very few examples (few-shot learning), successfully applying familiar functions to novel inputs. People can also compose concepts in complex ways that go beyond the provided demonstrations. Two additional experiments examined the assumptions and inductive biases that people make when solving these tasks, revealing three biases: mutual exclusivity, one-to-one mappings, and iconic concatenation. We discuss the implications for cognitive modeling and the potential for building machines with more human-like language learning capabilities.},\n  archiveprefix = {arxiv},\n  keywords = {Computer Science - Computation and Language},\n  file = {/Users/shanest/sync/library/Lake et al/2019/Lake et al. - 2019 - Human few-shot learning of compositional instructi.pdf;/Users/shanest/Zotero/storage/82PMFXNQ/1901.html}\n}\n\n
\n
\n\n\n
\n People learn in fast and flexible ways that have not been emulated by machines. Once a person learns a new verb \"dax,\" he or she can effortlessly understand how to \"dax twice,\" \"walk and dax,\" or \"dax vigorously.\" There have been striking recent improvements in machine learning for natural language processing, yet the best algorithms require vast amounts of experience and struggle to generalize new concepts in compositional ways. To better understand these distinctively human abilities, we study the compositional skills of people through language-like instruction learning tasks. Our results show that people can learn and use novel functional concepts from very few examples (few-shot learning), successfully applying familiar functions to novel inputs. People can also compose concepts in complex ways that go beyond the provided demonstrations. Two additional experiments examined the assumptions and inductive biases that people make when solving these tasks, revealing three biases: mutual exclusivity, one-to-one mappings, and iconic concatenation. We discuss the implications for cognitive modeling and the potential for building machines with more human-like language learning capabilities.\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n \n Compositional Generalization for Primitive Substitutions.\n \n \n \n \n\n\n \n Li, Y.; Zhao, L.; Wang, J.; and Hestness, J.\n\n\n \n\n\n\n In Inui, K.; Jiang, J.; Ng, V.; and Wan, X., editor(s), Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pages 4293–4302, Hong Kong, China, November 2019. Association for Computational Linguistics\n \n\n\n\n
\n\n\n\n \n \n \"CompositionalPaper\n  \n \n\n \n \n doi\n  \n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n  \n \n 1 download\n \n \n\n \n \n \n \n \n \n \n\n  \n \n \n\n\n\n
\n 
@inproceedings{liCompositionalGeneralizationPrimitive2019,\n  title = {Compositional {{Generalization}} for {{Primitive Substitutions}}},\n  booktitle = {Proceedings of the 2019 {{Conference}} on {{Empirical Methods}} in {{Natural Language Processing}} and the 9th {{International Joint Conference}} on {{Natural Language Processing}} ({{EMNLP-IJCNLP}})},\n  author = {Li, Yuanpeng and Zhao, Liang and Wang, Jianyu and Hestness, Joel},\n  editor = {Inui, Kentaro and Jiang, Jing and Ng, Vincent and Wan, Xiaojun},\n  year = {2019},\n  month = nov,\n  pages = {4293--4302},\n  publisher = {Association for Computational Linguistics},\n  address = {Hong Kong, China},\n  doi = {10.18653/v1/D19-1438},\n  url = {https://aclanthology.org/D19-1438},\n  urldate = {2024-03-19},\n  abstract = {Compositional generalization is a basic mechanism in human language learning, but current neural networks lack such ability. In this paper, we conduct fundamental research for encoding compositionality in neural networks. Conventional methods use a single representation for the input sentence, making it hard to apply prior knowledge of compositionality. In contrast, our approach leverages such knowledge with two representations, one generating attention maps, and the other mapping attended input words to output symbols. We reduce the entropy in each representation to improve generalization. Our experiments demonstrate significant improvements over the conventional methods in five NLP tasks including instruction learning and machine translation. In the SCAN domain, it boosts accuracies from 14.0\\% to 98.8\\% in Jump task, and from 92.0\\% to 99.7\\% in TurnLeft task. It also beats human performance on a few-shot learning task. We hope the proposed approach can help ease future research towards human-level compositional language learning.},\n  file = {/Users/shanest/sync/library/Li et al/2019/Li et al. - 2019 - Compositional Generalization for Primitive Substit.pdf}\n}\n\n
\n
\n\n\n
\n Compositional generalization is a basic mechanism in human language learning, but current neural networks lack such ability. In this paper, we conduct fundamental research for encoding compositionality in neural networks. Conventional methods use a single representation for the input sentence, making it hard to apply prior knowledge of compositionality. In contrast, our approach leverages such knowledge with two representations, one generating attention maps, and the other mapping attended input words to output symbols. We reduce the entropy in each representation to improve generalization. Our experiments demonstrate significant improvements over the conventional methods in five NLP tasks including instruction learning and machine translation. In the SCAN domain, it boosts accuracies from 14.0% to 98.8% in Jump task, and from 92.0% to 99.7% in TurnLeft task. It also beats human performance on a few-shot learning task. We hope the proposed approach can help ease future research towards human-level compositional language learning.\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n \n A Case for Deep Learning in Semantics: Response to Pater.\n \n \n \n \n\n\n \n Potts, C.\n\n\n \n\n\n\n Language, 95(1): e115-e124. 2019.\n \n\n\n\n
\n\n\n\n \n \n \"APaper\n  \n \n\n \n \n doi\n  \n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n  \n \n 1 download\n \n \n\n \n \n \n \n \n \n \n\n  \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n\n\n\n
\n 
@article{pottsCaseDeepLearning2019,\n  title = {A Case for Deep Learning in Semantics: {{Response}} to {{Pater}}},\n  shorttitle = {A Case for Deep Learning in Semantics},\n  author = {Potts, Christopher},\n  year = {2019},\n  journal = {Language},\n  volume = {95},\n  number = {1},\n  pages = {e115-e124},\n  publisher = {Linguistic Society of America},\n  issn = {1535-0665},\n  doi = {10.1353/lan.2019.0019},\n  url = {https://doi.org/10.1353/lan.2019.0019},\n  urldate = {2024-03-18},\n  abstract = {Pater's (2019) target article builds a persuasive case for establishing stronger ties between theoretical linguistics and connectionism (deep learning). This commentary extends his arguments to semantics, focusing in particular on issues of learning, compositionality, and lexical meaning.*},\n  keywords = {compositionality,connectionism,deep learning,lexical semantics,machine learning,position,semantics,survey},\n  file = {/Users/shanest/sync/library/Potts/2019/Potts - 2019 - A case for deep learning in semantics Response to2.pdf}\n}\n\n
\n
\n\n\n
\n Pater's (2019) target article builds a persuasive case for establishing stronger ties between theoretical linguistics and connectionism (deep learning). This commentary extends his arguments to semantics, focusing in particular on issues of learning, compositionality, and lexical meaning.*\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n \n Compositional Hierarchical Structure Evolves through Cultural Transmission: An Experimental Study.\n \n \n \n \n\n\n \n Saldana, C.; Kirby, S.; Truswell, R.; and Smith, K.\n\n\n \n\n\n\n Journal of Language Evolution, 4(2): 83–107. July 2019.\n \n\n\n\n
\n\n\n\n \n \n \"CompositionalPaper\n  \n \n\n \n \n doi\n  \n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n\n\n\n
\n 
@article{saldanaCompositionalHierarchicalStructure2019,\n  title = {Compositional {{Hierarchical Structure Evolves}} through {{Cultural Transmission}}: {{An Experimental Study}}},\n  shorttitle = {Compositional {{Hierarchical Structure Evolves}} through {{Cultural Transmission}}},\n  author = {Saldana, Carmen and Kirby, Simon and Truswell, Robert and Smith, Kenny},\n  year = {2019},\n  month = jul,\n  journal = {Journal of Language Evolution},\n  volume = {4},\n  number = {2},\n  pages = {83--107},\n  issn = {2058-458X},\n  doi = {10.1093/jole/lzz002},\n  url = {https://doi.org/10.1093/jole/lzz002},\n  urldate = {2024-04-05},\n  abstract = {Compositional hierarchical structure is a prerequisite for productive languages; it allows language learners to express and understand an infinity of meanings from finite sources (i.e., a lexicon and a grammar). Understanding how such structure evolved is central to evolutionary linguistics. Previous work combining artificial language learning and iterated learning techniques has shown how basic compositional structure can evolve from the trade-off between learnability and expressivity pressures at play in language transmission. In the present study we show, across two experiments, how the same mechanisms involved in the evolution of basic compositionality can also lead to the evolution of compositional hierarchical structure. We thus provide experimental evidence showing that cultural transmission allows advantages of compositional hierarchical structure in language learning and use to permeate language as a system of behaviour.}\n}\n\n
\n
\n\n\n
\n Compositional hierarchical structure is a prerequisite for productive languages; it allows language learners to express and understand an infinity of meanings from finite sources (i.e., a lexicon and a grammar). Understanding how such structure evolved is central to evolutionary linguistics. Previous work combining artificial language learning and iterated learning techniques has shown how basic compositional structure can evolve from the trade-off between learnability and expressivity pressures at play in language transmission. In the present study we show, across two experiments, how the same mechanisms involved in the evolution of basic compositionality can also lead to the evolution of compositional hierarchical structure. We thus provide experimental evidence showing that cultural transmission allows advantages of compositional hierarchical structure in language learning and use to permeate language as a system of behaviour.\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n \n Syntax and Compositionality in Animal Communication.\n \n \n \n \n\n\n \n Zuberbühler, K.\n\n\n \n\n\n\n Philosophical Transactions of the Royal Society B: Biological Sciences, 375(1789): 20190062. November 2019.\n \n\n\n\n
\n\n\n\n \n \n \"SyntaxPaper\n  \n \n\n \n \n doi\n  \n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n  \n \n 1 download\n \n \n\n \n \n \n \n \n \n \n\n  \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n\n\n\n
\n 
@article{zuberbuhlerSyntaxCompositionalityAnimal2019,\n  title = {Syntax and Compositionality in Animal Communication},\n  author = {Zuberb{\\"u}hler, Klaus},\n  year = {2019},\n  month = nov,\n  journal = {Philosophical Transactions of the Royal Society B: Biological Sciences},\n  volume = {375},\n  number = {1789},\n  pages = {20190062},\n  publisher = {Royal Society},\n  doi = {10.1098/rstb.2019.0062},\n  url = {https://royalsocietypublishing.org/doi/10.1098/rstb.2019.0062},\n  urldate = {2024-03-19},\n  abstract = {Syntax has been found in animal communication but only humans appear to have generative, hierarchically structured syntax. How did syntax evolve? I discuss three theories of evolutionary transition from animal to human syntax: computational capacity, structural flexibility and event perception. The computation hypothesis is supported by artificial grammar experiments consistently showing that only humans can learn linear stimulus sequences with an underlying hierarchical structure, a possible by-product of computationally powerful large brains. The structural flexibility hypothesis is supported by evidence of meaning-bearing combinatorial and permutational signal sequences in animals, with sometimes compositional features, but no evidence for generativity or hierarchical structure. Again, animals may be constrained by computational limits in short-term memory but possibly also by limits in articulatory control and social cognition. The event categorization hypothesis, finally, posits that humans are cognitively predisposed to analyse natural events by assigning agency and assessing how agents impact on patients, a propensity that is reflected by the basic syntactic units in all languages. Whether animals perceive natural events in the same way is largely unknown, although event perception may provide the cognitive grounding for syntax evolution. This article is part of the theme issue `What can animal communication teach us about human language?'},\n  keywords = {grammar,language evolution,meaning,permutation,primate communication,semantics},\n  file = {/Users/shanest/sync/library/Zuberbühler/2019/Zuberbühler - 2019 - Syntax and compositionality in animal communicatio.pdf}\n}\n
\n
\n\n\n
\n Syntax has been found in animal communication but only humans appear to have generative, hierarchically structured syntax. How did syntax evolve? I discuss three theories of evolutionary transition from animal to human syntax: computational capacity, structural flexibility and event perception. The computation hypothesis is supported by artificial grammar experiments consistently showing that only humans can learn linear stimulus sequences with an underlying hierarchical structure, a possible by-product of computationally powerful large brains. The structural flexibility hypothesis is supported by evidence of meaning-bearing combinatorial and permutational signal sequences in animals, with sometimes compositional features, but no evidence for generativity or hierarchical structure. Again, animals may be constrained by computational limits in short-term memory but possibly also by limits in articulatory control and social cognition. The event categorization hypothesis, finally, posits that humans are cognitively predisposed to analyse natural events by assigning agency and assessing how agents impact on patients, a propensity that is reflected by the basic syntactic units in all languages. Whether animals perceive natural events in the same way is largely unknown, although event perception may provide the cognitive grounding for syntax evolution. This article is part of the theme issue `What can animal communication teach us about human language?'\n
\n\n\n
\n\n\n\n\n\n
\n
\n\n
\n
\n  \n 2018\n \n \n (2)\n \n \n
\n
\n \n \n
\n \n\n \n \n \n \n \n \n Generalization without Systematicity: On the Compositional Skills of Sequence-to-Sequence Recurrent Networks.\n \n \n \n \n\n\n \n Lake, B.; and Baroni, M.\n\n\n \n\n\n\n In Proceedings of the 35th International Conference on Machine Learning, pages 2873–2882, July 2018. PMLR\n \n\n\n\n
\n\n\n\n \n \n \"GeneralizationPaper\n  \n \n\n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n  \n \n 8 downloads\n \n \n\n \n \n \n \n \n \n \n\n  \n \n \n\n\n\n
\n 
@inproceedings{lakeGeneralizationSystematicityCompositional2018,\n  title = {Generalization without {{Systematicity}}: {{On}} the {{Compositional Skills}} of {{Sequence-to-Sequence Recurrent Networks}}},\n  shorttitle = {Generalization without {{Systematicity}}},\n  booktitle = {Proceedings of the 35th {{International Conference}} on {{Machine Learning}}},\n  author = {Lake, Brenden and Baroni, Marco},\n  year = {2018},\n  month = jul,\n  pages = {2873--2882},\n  publisher = {PMLR},\n  issn = {2640-3498},\n  url = {https://proceedings.mlr.press/v80/lake18a.html},\n  urldate = {2024-03-18},\n  abstract = {Humans can understand and produce new utterances effortlessly, thanks to their compositional skills. Once a person learns the meaning of a new verb "dax," he or she can immediately understand the meaning of "dax twice" or "sing and dax." In this paper, we introduce the SCAN domain, consisting of a set of simple compositional navigation commands paired with the corresponding action sequences. We then test the zero-shot generalization capabilities of a variety of recurrent neural networks (RNNs) trained on SCAN with sequence-to-sequence methods. We find that RNNs can make successful zero-shot generalizations when the differences between training and test commands are small, so that they can apply "mix-and-match" strategies to solve the task. However, when generalization requires systematic compositional skills (as in the "dax" example above), RNNs fail spectacularly. We conclude with a proof-of-concept experiment in neural machine translation, suggesting that lack of systematicity might be partially responsible for neural networks' notorious training data thirst.},\n  langid = {english},\n  file = {/Users/shanest/sync/library/Lake_Baroni/2018/Lake and Baroni - 2018 - Generalization without Systematicity On the Compo2.pdf;/Users/shanest/sync/library/Lake_Baroni/2018/Lake and Baroni - 2018 - Generalization without Systematicity On the Compo3.pdf}\n}\n\n
\n
\n\n\n
\n Humans can understand and produce new utterances effortlessly, thanks to their compositional skills. Once a person learns the meaning of a new verb \"dax,\" he or she can immediately understand the meaning of \"dax twice\" or \"sing and dax.\" In this paper, we introduce the SCAN domain, consisting of a set of simple compositional navigation commands paired with the corresponding action sequences. We then test the zero-shot generalization capabilities of a variety of recurrent neural networks (RNNs) trained on SCAN with sequence-to-sequence methods. We find that RNNs can make successful zero-shot generalizations when the differences between training and test commands are small, so that they can apply \"mix-and-match\" strategies to solve the task. However, when generalization requires systematic compositional skills (as in the \"dax\" example above), RNNs fail spectacularly. We conclude with a proof-of-concept experiment in neural machine translation, suggesting that lack of systematicity might be partially responsible for neural networks' notorious training data thirst.\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n \n Emergence of Grounded Compositional Language in Multi-Agent Populations.\n \n \n \n \n\n\n \n Mordatch, I.; and Abbeel, P.\n\n\n \n\n\n\n In Proceedings of the AAAI Conference on Artificial Intelligence, volume 32, April 2018. \n \n\n\n\n
\n\n\n\n \n \n \"EmergencePaper\n  \n \n\n \n \n doi\n  \n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n  \n \n 1 download\n \n \n\n \n \n \n \n \n \n \n\n  \n \n \n \n \n\n\n\n
\n 
@inproceedings{mordatchEmergenceGroundedCompositional2018,\n  title = {Emergence of {{Grounded Compositional Language}} in {{Multi-Agent Populations}}},\n  booktitle = {Proceedings of the {{AAAI Conference}} on {{Artificial Intelligence}}},\n  author = {Mordatch, Igor and Abbeel, Pieter},\n  year = {2018},\n  month = apr,\n  volume = {32},\n  doi = {10.1609/aaai.v32i1.11492},\n  url = {https://ojs.aaai.org/index.php/AAAI/article/view/11492},\n  urldate = {2024-03-19},\n  abstract = {By capturing statistical patterns in large corpora, machine learning has enabled significant advances in natural language processing, including in machine translation, question answering, and sentiment analysis. However, for agents to intelligently interact with humans, simply capturing the statistical patterns is insufficient. In this paper we investigate if, and how, grounded compositional language can emerge as a means to achieve goals in multi-agent populations. Towards this end, we propose a multi-agent learning environment and learning methods that bring about emergence of a basic compositional language. This language is represented as streams of abstract discrete symbols uttered by agents over time, but nonetheless has a coherent structure that possesses a defined vocabulary and syntax. We also observe emergence of non-verbal communication such as pointing and guiding when language communication is unavailable.},\n  copyright = {Copyright (c)},\n  langid = {english},\n  keywords = {Language Emergence},\n  file = {/Users/shanest/sync/library/Mordatch_Abbeel/2018/Mordatch and Abbeel - 2018 - Emergence of Grounded Compositional Language in Mu.pdf;/Users/shanest/Zotero/storage/ZMYMA7FP/Mordatch and Abbeel - 2018 - Emergence of Grounded Compositional Language in Mu.pdf}\n}\n\n
\n
\n\n\n
\n By capturing statistical patterns in large corpora, machine learning has enabled significant advances in natural language processing, including in machine translation, question answering, and sentiment analysis. However, for agents to intelligently interact with humans, simply capturing the statistical patterns is insufficient. In this paper we investigate if, and how, grounded compositional language can emerge as a means to achieve goals in multi-agent populations. Towards this end, we propose a multi-agent learning environment and learning methods that bring about emergence of a basic compositional language. This language is represented as streams of abstract discrete symbols uttered by agents over time, but nonetheless has a coherent structure that possesses a defined vocabulary and syntax. We also observe emergence of non-verbal communication such as pointing and guiding when language communication is unavailable.\n
\n\n\n
\n\n\n\n\n\n
\n
\n\n
\n
\n  \n 2017\n \n \n (1)\n \n \n
\n
\n \n \n
\n \n\n \n \n \n \n \n \n Self-Assembling Games.\n \n \n \n \n\n\n \n Barrett, J. A.; and Skyrms, B.\n\n\n \n\n\n\n The British Journal for the Philosophy of Science, 68(2): 329–353. June 2017.\n \n\n\n\n
\n\n\n\n \n \n \"Self-AssemblingPaper\n  \n \n\n \n \n doi\n  \n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n\n\n\n
\n 
@article{barrettSelfassemblingGames2017,\n  title = {Self-Assembling {{Games}}},\n  author = {Barrett, Jeffrey A. and Skyrms, Brian},\n  year = {2017},\n  month = jun,\n  journal = {The British Journal for the Philosophy of Science},\n  volume = {68},\n  number = {2},\n  pages = {329--353},\n  publisher = {The University of Chicago Press},\n  issn = {0007-0882},\n  doi = {10.1093/bjps/axv043},\n  url = {https://www.journals.uchicago.edu/doi/10.1093/bjps/axv043},\n  urldate = {2024-04-05},\n  abstract = {We consider how cue-reading, sensory-manipulation, and signalling games may initially evolve from ritualized decisions and how more complex games may evolve from simpler games by polymerization, template transfer, and modular composition. Modular composition is a process that combines simpler games into more complex games. Template transfer, a process by which a game is appropriated to a context other than the one in which it initially evolved, is one mechanism for modular composition. And polymerization is a particularly salient example of modular composition where simpler games evolve to form more complex chains. We also consider how the evolution of new capacities by modular composition may be more efficient than evolving those capacities from basic decisions. 1.~ Introduction 2.~ The Assembly of Decisions into Games: Cue-Reading, Sensory-Manipulation, and Signalling 3.~ Polymerization 4.~ Template Transfer 5.~ Modular Composition 6.~ Relative Efficiency of Template Transfer 7.~ Discussion}\n}\n\n
\n
\n\n\n
\n We consider how cue-reading, sensory-manipulation, and signalling games may initially evolve from ritualized decisions and how more complex games may evolve from simpler games by polymerization, template transfer, and modular composition. Modular composition is a process that combines simpler games into more complex games. Template transfer, a process by which a game is appropriated to a context other than the one in which it initially evolved, is one mechanism for modular composition. And polymerization is a particularly salient example of modular composition where simpler games evolve to form more complex chains. We also consider how the evolution of new capacities by modular composition may be more efficient than evolving those capacities from basic decisions. 1.  Introduction 2.  The Assembly of Decisions into Games: Cue-Reading, Sensory-Manipulation, and Signalling 3.  Polymerization 4.  Template Transfer 5.  Modular Composition 6.  Relative Efficiency of Template Transfer 7.  Discussion\n
\n\n\n
\n\n\n\n\n\n
\n
\n\n
\n
\n  \n 2016\n \n \n (4)\n \n \n
\n
\n \n \n
\n \n\n \n \n \n \n \n The Evolution of Compositionality in Signaling Games.\n \n \n \n\n\n \n Franke, M.\n\n\n \n\n\n\n Journal of Logic, Language and Information. 2016.\n \n\n\n\n
\n\n\n\n \n\n \n \n doi\n  \n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n \n \n \n \n \n \n\n\n\n
\n 
@article{frankeEvolutionCompositionalitySignaling2016,\n  title = {The {{Evolution}} of {{Compositionality}} in {{Signaling Games}}},\n  author = {Franke, Michael},\n  year = {2016},\n  journal = {Journal of Logic, Language and Information},\n  doi = {10.1007/s10849-015-9232-5},\n  abstract = {Compositionality is a key design feature of human language: the meaning of complex expressions is, for the most part, systematically constructed from the meanings of its parts and their manner of composition. This paper demonstrates that rudimentary forms of compositional communicative behavior can emerge from a variant of reinforcement learning applied to signaling games. This helps explain how compositionality could have emerged gradually: if unsophisticated agents can evolve prevalent dispositions to communicate compositional-like, there is a direct evolution- ary benefit for adaptations that exploit the systematicity in form-meaning mappings more rigorously.},\n  keywords = {Compositionality,Reinforcement learning,Signaling games},\n  file = {/Users/shanest/sync/library/Franke/2016/Franke - 2016 - The Evolution of Compositionality in Signaling Gam.pdf}\n}\n\n
\n
\n\n\n
\n Compositionality is a key design feature of human language: the meaning of complex expressions is, for the most part, systematically constructed from the meanings of its parts and their manner of composition. This paper demonstrates that rudimentary forms of compositional communicative behavior can emerge from a variant of reinforcement learning applied to signaling games. This helps explain how compositionality could have emerged gradually: if unsophisticated agents can evolve prevalent dispositions to communicate compositional-like, there is a direct evolution- ary benefit for adaptations that exploit the systematicity in form-meaning mappings more rigorously.\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n Formal Monkey Linguistics.\n \n \n \n\n\n \n Schlenker, P.; Chemla, E.; Schel, A. M; Fuller, J.; Gautier, J.; Kuhn, J.; Veselinović, D.; Arnold, K.; Cäsar, C.; Keenan, S.; Lemasson, A.; Ouattara, K.; Ryder, R.; and Zuberbühler, K.\n\n\n \n\n\n\n Theoretical Linguistics, 42(1-2): 1–90. 2016.\n \n\n\n\n
\n\n\n\n \n\n \n \n doi\n  \n \n\n \n link\n  \n \n\n bibtex\n \n\n \n\n \n  \n \n 3 downloads\n \n \n\n \n \n \n \n \n \n \n\n  \n \n \n\n\n\n
\n 
@article{schlenkerFormalMonkeyLinguistics2016,\n  title = {Formal Monkey Linguistics},\n  author = {Schlenker, Philippe and Chemla, Emmanuel and Schel, Anne M and Fuller, James and Gautier, Jean-Pierre and Kuhn, Jeremy and Veselinovi{\\'c}, Dunja and Arnold, Kate and C{\\"a}sar, Cristiane and Keenan, Sumir and Lemasson, Alban and Ouattara, Karim and Ryder, Robin and Zuberb{\\"u}hler, Klaus},\n  year = {2016},\n  journal = {Theoretical Linguistics},\n  volume = {42},\n  number = {1-2},\n  pages = {1--90},\n  doi = {10.1515/tl-2016-0001},\n  file = {/Users/shanest/sync/library/Schlenker et al/2016/Schlenker et al. - 2016 - Formal monkey linguistics2.pdf}\n}\n\n
\n
\n\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n Compositionality and Competition in Monkey Alert Calls.\n \n \n \n\n\n \n Steinert-Threlkeld, S.\n\n\n \n\n\n\n Theoretical Linguistics, 42(1-2): 159–171. 2016.\n \n\n\n\n
\n\n\n\n \n\n \n \n doi\n  \n \n\n \n link\n  \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n\n\n\n
\n 
@article{steinert-threlkeldCompositionalityCompetitionMonkey2016,\n  title = {Compositionality and Competition in Monkey Alert Calls},\n  author = {{Steinert-Threlkeld}, Shane},\n  year = {2016},\n  journal = {Theoretical Linguistics},\n  volume = {42},\n  number = {1-2},\n  pages = {159--171},\n  doi = {10.1515/tl-2016-0009},\n  file = {/Users/shanest/sync/library/Steinert-Threlkeld/2016/Steinert-Threlkeld - 2016 - Compositionality and competition in monkey alert c.pdf}\n}\n\n
\n
\n\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n Compositional Signaling in a Complex World.\n \n \n \n\n\n \n Steinert-Threlkeld, S.\n\n\n \n\n\n\n Journal of Logic, Language and Information, 25(3): 379–397. 2016.\n \n\n\n\n
\n\n\n\n \n\n \n \n doi\n  \n \n\n \n link\n  \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n\n\n\n
\n 
@article{steinert-threlkeldCompositionalSignalingComplex2016,\n  title = {Compositional {{Signaling}} in a {{Complex World}}},\n  author = {{Steinert-Threlkeld}, Shane},\n  year = {2016},\n  journal = {Journal of Logic, Language and Information},\n  volume = {25},\n  number = {3},\n  pages = {379--397},\n  doi = {10.1007/s10849-016-9236-9},\n  file = {/Users/shanest/sync/library/Steinert-Threlkeld/2016/Steinert-Threlkeld - 2016 - Compositional Signaling in a Complex World.pdf}\n}\n\n
\n
\n\n\n\n
\n\n\n\n\n\n
\n
\n\n
\n
\n  \n 2015\n \n \n (1)\n \n \n
\n
\n \n \n
\n \n\n \n \n \n \n \n \n Compression and Communication in the Cultural Evolution of Linguistic Structure.\n \n \n \n \n\n\n \n Kirby, S.; Tamariz, M.; Cornish, H.; and Smith, K.\n\n\n \n\n\n\n Cognition, 141: 87–102. 2015.\n \n\n\n\n
\n\n\n\n \n \n \"CompressionPaper\n  \n \n\n \n \n doi\n  \n \n\n \n link\n  \n \n\n bibtex\n \n\n \n\n \n  \n \n 12 downloads\n \n \n\n \n \n \n \n \n \n \n\n  \n \n \n \n \n\n\n\n
\n 
@article{kirbyCompressionCommunicationCultural2015,\n  title = {Compression and Communication in the Cultural Evolution of Linguistic Structure},\n  author = {Kirby, Simon and Tamariz, Monica and Cornish, Hannah and Smith, Kenny},\n  year = {2015},\n  journal = {Cognition},\n  volume = {141},\n  pages = {87--102},\n  publisher = {Elsevier B.V.},\n  issn = {00100277},\n  doi = {10.1016/j.cognition.2015.03.016},\n  url = {http://linkinghub.elsevier.com/retrieve/pii/S0010027715000815},\n  keywords = {cultural transmission},\n  file = {/Users/shanest/sync/library/Kirby et al/2015/Kirby et al. - 2015 - Compression and communication in the cultural evol.pdf}\n}\n\n
\n
\n\n\n\n
\n\n\n\n\n\n
\n
\n\n
\n
\n  \n 2013\n \n \n (2)\n \n \n
\n
\n \n \n
\n \n\n \n \n \n \n \n \n The Dance Language and Orientation of Bees.\n \n \n \n \n\n\n \n von Frisch, K.\n\n\n \n\n\n\n In The Dance Language and Orientation of Bees. Harvard University Press, October 2013.\n \n\n\n\n
\n\n\n\n \n \n \"ThePaper\n  \n \n\n \n \n doi\n  \n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n \n \n \n \n \n \n \n \n \n \n\n\n\n
\n 
@incollection{frischDanceLanguageOrientation2013,\n  title = {The {{Dance Language}} and {{Orientation}} of {{Bees}}},\n  booktitle = {The {{Dance Language}} and {{Orientation}} of {{Bees}}},\n  author = {von Frisch, Karl},\n  year = {2013},\n  month = oct,\n  publisher = {Harvard University Press},\n  doi = {10.4159/harvard.9780674418776},\n  url = {https://www.degruyter.com/document/doi/10.4159/harvard.9780674418776/html},\n  urldate = {2024-03-19},\n  abstract = {Until his death in 1982, Karl von Frisch was the world's most renowned authority on bees. The Dance Language and Orientation of Bees is his masterwork--the culmination of more than fifty years of research. Now available for the first time in paperback, it describes in non-technical language what he discovered in a lifetime of study about honeybees--their methods of orientation, their sensory faculties, and their remarkable ability to communicate with one another. Thomas Seeley's new foreword traces the revolutionary effects of von Frisch's work, not just for the study of bees, but for all subsequent research in animal behavior. This new paperback edition also includes an "Appreciation" of von Frisch by the distinguished biologist Martin Lindauer, who was Frisch's prot{\\'e}g{\\'e} and later his colleague and friend.},\n  copyright = {De Gruyter expressly reserves the right to use all content for commercial text and data mining within the meaning of Section 44b of the German Copyright Act.},\n  isbn = {978-0-674-41877-6},\n  langid = {english},\n  keywords = {Animal communication.,Animal orientation.,Bees.,Insects -- Behavior.,SCIENCE / Life Sciences / Biology},\n  file = {/Users/shanest/Zotero/storage/G6MVI8DU/Frisch - 2013 - The Dance Language and Orientation of Bees.pdf}\n}\n\n
\n
\n\n\n
\n Until his death in 1982, Karl von Frisch was the world's most renowned authority on bees. The Dance Language and Orientation of Bees is his masterwork–the culmination of more than fifty years of research. Now available for the first time in paperback, it describes in non-technical language what he discovered in a lifetime of study about honeybees–their methods of orientation, their sensory faculties, and their remarkable ability to communicate with one another. Thomas Seeley's new foreword traces the revolutionary effects of von Frisch's work, not just for the study of bees, but for all subsequent research in animal behavior. This new paperback edition also includes an \"Appreciation\" of von Frisch by the distinguished biologist Martin Lindauer, who was Frisch's protégé and later his colleague and friend.\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n Linguistic Structure Is an Evolutionary Trade-off between Simplicity and Expressivity.\n \n \n \n\n\n \n Smith, K.; Tamariz, M.; and Kirby, S.\n\n\n \n\n\n\n Proceedings of the 35th Annual Meeting of the Cognitive Science Society,1348–1353. 2013.\n \n\n\n\n
\n\n\n\n \n\n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n \n \n \n \n \n \n \n \n\n\n\n
\n 
@article{smithLinguisticStructureEvolutionary2013,\n  title = {Linguistic Structure Is an Evolutionary Trade-off between Simplicity and Expressivity},\n  author = {Smith, Kenny and Tamariz, Monica and Kirby, Simon},\n  year = {2013},\n  journal = {Proceedings of the 35th Annual Meeting of the Cognitive Science Society},\n  pages = {1348--1353},\n  abstract = {Language exhibits structure: a species-unique system for expressing complex meanings using complex forms. We present a review of modelling and experimental literature on the evolution of structure which suggests that structure is a cultural adaptation in response to pressure for expressivity (arising during communication) and compressibility (arising during learning), and test this hypothesis using a new Bayesian iterated learning model. We conclude that linguistic structure can and should be explained as a consequence of cultural evolution in response to these two pressures.},\n  keywords = {cultural evolution,language,learning,structure},\n  file = {/Users/shanest/sync/library/Smith et al/2013/Smith et al. - 2013 - Linguistic structure is an evolutionary trade-off .pdf}\n}\n\n
\n
\n\n\n
\n Language exhibits structure: a species-unique system for expressing complex meanings using complex forms. We present a review of modelling and experimental literature on the evolution of structure which suggests that structure is a cultural adaptation in response to pressure for expressivity (arising during communication) and compressibility (arising during learning), and test this hypothesis using a new Bayesian iterated learning model. We conclude that linguistic structure can and should be explained as a consequence of cultural evolution in response to these two pressures.\n
\n\n\n
\n\n\n\n\n\n
\n
\n\n
\n
\n  \n 2010\n \n \n (3)\n \n \n
\n
\n \n \n
\n \n\n \n \n \n \n \n Compositionality I: Definitions and Variants.\n \n \n \n\n\n \n Pagin, P.; and Westerståhl, D.\n\n\n \n\n\n\n Philosophy Compass, 5(3): 250–264. 2010.\n \n\n\n\n
\n\n\n\n \n\n \n \n doi\n  \n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n\n\n\n
\n 
@article{paginCompositionalityDefinitionsVariants2010,\n  title = {Compositionality {{I}}: {{Definitions}} and {{Variants}}.},\n  author = {Pagin, Peter and Westerst{\\aa}hl, Dag},\n  year = {2010},\n  journal = {Philosophy Compass},\n  volume = {5},\n  number = {3},\n  pages = {250--264},\n  doi = {10.1111/j.1747-9991.2009.00228.x},\n  abstract = {This is the first part of a two-part article on semantic compositionality, that is, the principle that the meaning of a complex expression is determined by the meanings of its parts and the way they are put together. Here we provide a brief historical background, a formal framework for syntax and semantics, precise definitions, and a survey of variants of compositionality. Stronger and weaker forms are distinguished, as well as generalized forms that cover extra-linguistic context dependence as well as linguistic context dependence. In the second article, we survey arguments for and arguments against the claim that natural languages are compositional, and consider some problem cases. It will be referred to as Part II.},\n  file = {/Users/shanest/sync/library/Pagin_Westerstr a hl/2010/Pagin and Westerstr a hl - 2010 - Compositionality I Definitions and Variants..pdf}\n}\n\n
\n
\n\n\n
\n This is the first part of a two-part article on semantic compositionality, that is, the principle that the meaning of a complex expression is determined by the meanings of its parts and the way they are put together. Here we provide a brief historical background, a formal framework for syntax and semantics, precise definitions, and a survey of variants of compositionality. Stronger and weaker forms are distinguished, as well as generalized forms that cover extra-linguistic context dependence as well as linguistic context dependence. In the second article, we survey arguments for and arguments against the claim that natural languages are compositional, and consider some problem cases. It will be referred to as Part II.\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n Compositionality II: Arguments and Problems.\n \n \n \n\n\n \n Pagin, P.; and Westerståhl, D.\n\n\n \n\n\n\n Philosophy Compass, 5(3): 265–282. 2010.\n \n\n\n\n
\n\n\n\n \n\n \n \n doi\n  \n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n\n\n\n
\n 
@article{paginCompositionalityIIArguments2010,\n  title = {Compositionality {{II}}: {{Arguments}} and {{Problems}}.},\n  author = {Pagin, Peter and Westerst{\\aa}hl, Dag},\n  year = {2010},\n  journal = {Philosophy Compass},\n  volume = {5},\n  number = {3},\n  pages = {265--282},\n  doi = {10.1111/j.1747-9991.2009.00229.x},\n  abstract = {This is the second part of a two-part article on compositionality, i.e. the principle that the meaning of a complex expression is determined by the meanings of its parts and the way they are put together. In the first, Pagin and Westerst{\\aa}hl (2010), we provide a general historical background, a formal framework, definitions, and a survey of variants of compositionality. It will be referred to as Part I. Here we discuss arguments for and against the claim that natural languages have a compositional semantics. We also discuss some problem cases, including belief reports, quotation, idioms, and ambiguity.},\n  file = {/Users/shanest/sync/library/Pagin_Westerstr a hl/2010/Pagin and Westerstr a hl - 2010 - Compositionality II Arguments and Problems..pdf}\n}\n\n
\n
\n\n\n
\n This is the second part of a two-part article on compositionality, i.e. the principle that the meaning of a complex expression is determined by the meanings of its parts and the way they are put together. In the first, Pagin and Westerståhl (2010), we provide a general historical background, a formal framework, definitions, and a survey of variants of compositionality. It will be referred to as Part I. Here we discuss arguments for and against the claim that natural languages have a compositional semantics. We also discuss some problem cases, including belief reports, quotation, idioms, and ambiguity.\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n Signals: Evolution, Learning, and Information.\n \n \n \n\n\n \n Skyrms, B.\n\n\n \n\n\n\n Oxford University Press, 2010.\n \n\n\n\n
\n\n\n\n \n\n \n\n \n link\n  \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n \n \n\n\n\n
\n 
@book{skyrmsSignalsEvolutionLearning2010a,\n  title = {Signals: {{Evolution}}, {{Learning}}, and {{Information}}},\n  shorttitle = {Signals},\n  author = {Skyrms, Brian},\n  year = {2010},\n  publisher = {Oxford University Press},\n  keywords = {Meaning}\n}\n\n
\n
\n\n\n\n
\n\n\n\n\n\n
\n
\n\n
\n
\n  \n 2009\n \n \n (1)\n \n \n
\n
\n \n \n
\n \n\n \n \n \n \n \n The Evolution of Coding in Signaling Games.\n \n \n \n\n\n \n Barrett, J. A\n\n\n \n\n\n\n Theory and Decision, 67(2): 223–237. 2009.\n \n\n\n\n
\n\n\n\n \n\n \n \n doi\n  \n \n\n \n link\n  \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n \n \n \n \n \n \n\n\n\n
\n 
@article{barrettEvolutionCodingSignaling2009,\n  title = {The {{Evolution}} of {{Coding}} in {{Signaling Games}}},\n  author = {Barrett, Jeffrey A},\n  year = {2009},\n  journal = {Theory and Decision},\n  volume = {67},\n  number = {2},\n  pages = {223--237},\n  doi = {10.1007/s11238-007-9064-0},\n  urldate = {2013-05-25},\n  isbn = {1123800790},\n  keywords = {evolution of language,evolutionary game theory,signaling},\n  file = {/Users/shanest/sync/library/Barrett/2009/Barrett - 2009 - The Evolution of Coding in Signaling Games.pdf}\n}\n\n
\n
\n\n\n\n
\n\n\n\n\n\n
\n
\n\n
\n
\n  \n 2008\n \n \n (1)\n \n \n
\n
\n \n \n
\n \n\n \n \n \n \n \n Cumulative Cultural Evolution in the Laboratory: An Experimental Approach to the Origins of Structure in Human Language.\n \n \n \n\n\n \n Kirby, S.; Cornish, H.; and Smith, K.\n\n\n \n\n\n\n Proceedings of the National Academy of Sciences of the United States of America, 105(31): 10681–10686. 2008.\n \n\n\n\n
\n\n\n\n \n\n \n \n doi\n  \n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n\n\n\n
\n 
@article{kirbyCumulativeCulturalEvolution2008,\n  title = {Cumulative Cultural Evolution in the Laboratory: An Experimental Approach to the Origins of Structure in Human Language.},\n  author = {Kirby, Simon and Cornish, Hannah and Smith, Kenny},\n  year = {2008},\n  journal = {Proceedings of the National Academy of Sciences of the United States of America},\n  volume = {105},\n  number = {31},\n  pages = {10681--10686},\n  issn = {0027-8424},\n  doi = {10.1073/pnas.0707835105},\n  abstract = {We introduce an experimental paradigm for studying the cumulative cultural evolution of language. In doing so we provide the first experimental validation for the idea that cultural transmission can lead to the appearance of design without a designer. Our experiments involve the iterated learning of artificial languages by human participants. We show that languages transmitted culturally evolve in such a way as to maximize their own transmissibility: over time, the languages in our experiments become easier to learn and increasingly structured. Furthermore, this structure emerges purely as a consequence of the transmission of language over generations, without any intentional design on the part of individual language learners. Previous computational and mathematical models suggest that iterated learning provides an explanation for the structure of human language and link particular aspects of linguistic structure with particular constraints acting on language during its transmission. The experimental work presented here shows that the predictions of these models, and models of cultural evolution more generally, can be tested in the laboratory.},\n  isbn = {1068110686},\n  pmid = {18667697},\n  file = {/Users/shanest/sync/library/Kirby et al/2008/Kirby et al. - 2008 - Cumulative cultural evolution in the laboratory a.pdf}\n}\n\n
\n
\n\n\n
\n We introduce an experimental paradigm for studying the cumulative cultural evolution of language. In doing so we provide the first experimental validation for the idea that cultural transmission can lead to the appearance of design without a designer. Our experiments involve the iterated learning of artificial languages by human participants. We show that languages transmitted culturally evolve in such a way as to maximize their own transmissibility: over time, the languages in our experiments become easier to learn and increasingly structured. Furthermore, this structure emerges purely as a consequence of the transmission of language over generations, without any intentional design on the part of individual language learners. Previous computational and mathematical models suggest that iterated learning provides an explanation for the structure of human language and link particular aspects of linguistic structure with particular constraints acting on language during its transmission. The experimental work presented here shows that the predictions of these models, and models of cultural evolution more generally, can be tested in the laboratory.\n
\n\n\n
\n\n\n\n\n\n
\n
\n\n
\n
\n  \n 2005\n \n \n (1)\n \n \n
\n
\n \n \n
\n \n\n \n \n \n \n \n \n Language as an Evolutionary System.\n \n \n \n \n\n\n \n Brighton, H.; Smith, K.; and Kirby, S.\n\n\n \n\n\n\n Physics of Life Reviews, 2(3): 177–226. September 2005.\n \n\n\n\n
\n\n\n\n \n \n \"LanguagePaper\n  \n \n\n \n \n doi\n  \n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n  \n \n 3 downloads\n \n \n\n \n \n \n \n \n \n \n\n  \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n\n\n\n
\n 
@article{brightonLanguageEvolutionarySystem2005,\n  title = {Language as an Evolutionary System},\n  author = {Brighton, Henry and Smith, Kenny and Kirby, Simon},\n  year = {2005},\n  month = sep,\n  journal = {Physics of Life Reviews},\n  volume = {2},\n  number = {3},\n  pages = {177--226},\n  issn = {1571-0645},\n  doi = {10.1016/j.plrev.2005.06.001},\n  url = {https://www.sciencedirect.com/science/article/pii/S1571064505000229},\n  urldate = {2024-03-18},\n  abstract = {John Maynard Smith and E{\\"o}rs Szathm{\\'a}ry argued that human language signified the eighth major transition in evolution: human language marked a new form of information transmission from one generation to another [Maynard Smith J, Szathm{\\'a}ry E. The major transitions in evolution. Oxford: Oxford Univ. Press; 1995]. According to this view language codes cultural information and as such forms the basis for the evolution of complexity in human culture. In this article we develop the theory that language also codes information in another sense: languages code information on their own structure. As a result, languages themselves provide information that influences their own survival. To understand the consequences of this theory we discuss recent computational models of linguistic evolution. Linguistic evolution is the process by which languages themselves evolve. This article draws together this recent work on linguistic evolution and highlights the significance of this process in understanding the evolution of linguistic complexity. Our conclusions are that: (1) the process of linguistic transmission constitutes the basis for an evolutionary system, and (2), that this evolutionary system is only superficially comparable to the process of biological evolution.},\n  keywords = {adaptation,Adaptation,artificial life,Artificial life,culture,Culture,evolution,Evolution,language,Language,replication,Replication},\n  file = {/Users/shanest/sync/library/Brighton et al/2005/Brighton et al. - 2005 - Language as an evolutionary system.pdf;/Users/shanest/Zotero/storage/NFULIKPP/S1571064505000229.html}\n}\n\n
\n
\n\n\n
\n John Maynard Smith and Eörs Szathmáry argued that human language signified the eighth major transition in evolution: human language marked a new form of information transmission from one generation to another [Maynard Smith J, Szathmáry E. The major transitions in evolution. Oxford: Oxford Univ. Press; 1995]. According to this view language codes cultural information and as such forms the basis for the evolution of complexity in human culture. In this article we develop the theory that language also codes information in another sense: languages code information on their own structure. As a result, languages themselves provide information that influences their own survival. To understand the consequences of this theory we discuss recent computational models of linguistic evolution. Linguistic evolution is the process by which languages themselves evolve. This article draws together this recent work on linguistic evolution and highlights the significance of this process in understanding the evolution of linguistic complexity. Our conclusions are that: (1) the process of linguistic transmission constitutes the basis for an evolutionary system, and (2), that this evolutionary system is only superficially comparable to the process of biological evolution.\n
\n\n\n
\n\n\n\n\n\n
\n
\n\n
\n
\n  \n 1988\n \n \n (1)\n \n \n
\n
\n \n \n
\n \n\n \n \n \n \n \n Connectionism and Cognitive Architecture.\n \n \n \n\n\n \n Fodor, J.; and Pylyshyn, Z. W.\n\n\n \n\n\n\n Cognition, 28(1-2): 3–71. 1988.\n \n\n\n\n
\n\n\n\n \n\n \n \n doi\n  \n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n\n\n\n
\n 
@article{fodorConnectionismCognitiveArchitecture1988,\n  title = {Connectionism and {{Cognitive Architecture}}},\n  author = {Fodor, Jerry and Pylyshyn, Zenon W.},\n  year = {1988},\n  journal = {Cognition},\n  volume = {28},\n  number = {1-2},\n  pages = {3--71},\n  doi = {10.1016/0010-0277(88)90031-5},\n  urldate = {2011-10-31},\n  abstract = {This paper explores differences between Connectionist proposals for cognitive architecture and the sorts of models that have traditionally been assumed in cognitive science. We claim that the major distinction is that, while both Connectionist and Classical architectures postulate representational mental states, the latter but not the former are committed to a symbol-level of representation, or to a `language of thought': i.e., to representational states that have combinatorial syntactic and semantic structure. Several arguments for combinatorial structure in mental representations are then reviewed. These include arguments based on the `systematicity' of mental representation: i.e., on the fact that cognitive capacities always exhibit certain symmetries, so that the ability to entertain a given thought implies the ability to entertain thoughts with semantically related contents. We claim that such arguments make a powerful case that mind/brain architecture is not Connectionist at the cognitive level. We then consider the possibility that Connectionism may provide an account of the neural (or `abstract neurological') structures in which Classical cognitive architecture is implemented. We survey a number of the standard arguments that have been offered in favor of Connectionism, and conclude that they are coherent only on this interpretation.},\n  file = {/Users/shanest/sync/library/Fodor_Pylyshyn/1988/Fodor and Pylyshyn - 1988 - Connectionism and Cognitive Architecture.pdf}\n}\n\n
\n
\n\n\n
\n This paper explores differences between Connectionist proposals for cognitive architecture and the sorts of models that have traditionally been assumed in cognitive science. We claim that the major distinction is that, while both Connectionist and Classical architectures postulate representational mental states, the latter but not the former are committed to a symbol-level of representation, or to a `language of thought': i.e., to representational states that have combinatorial syntactic and semantic structure. Several arguments for combinatorial structure in mental representations are then reviewed. These include arguments based on the `systematicity' of mental representation: i.e., on the fact that cognitive capacities always exhibit certain symmetries, so that the ability to entertain a given thought implies the ability to entertain thoughts with semantically related contents. We claim that such arguments make a powerful case that mind/brain architecture is not Connectionist at the cognitive level. We then consider the possibility that Connectionism may provide an account of the neural (or `abstract neurological') structures in which Classical cognitive architecture is implemented. We survey a number of the standard arguments that have been offered in favor of Connectionism, and conclude that they are coherent only on this interpretation.\n
\n\n\n
\n\n\n\n\n\n
\n
\n\n
\n
\n  \n 1975\n \n \n (1)\n \n \n
\n
\n \n \n
\n \n\n \n \n \n \n \n \n Honey Bee Recruitment: The Dance-Language Controversy.\n \n \n \n \n\n\n \n Gould, J. L.\n\n\n \n\n\n\n Science, 189(4204): 685–693. 1975.\n \n\n\n\n
\n\n\n\n \n \n \"HoneyPaper\n  \n \n\n \n\n \n link\n  \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n\n\n\n
\n 
@article{gouldHoneyBeeRecruitment1975,\n  title = {Honey {{Bee Recruitment}}: {{The Dance-Language Controversy}}},\n  shorttitle = {Honey {{Bee Recruitment}}},\n  author = {Gould, James L.},\n  year = {1975},\n  journal = {Science},\n  volume = {189},\n  number = {4204},\n  eprint = {1740672},\n  eprinttype = {jstor},\n  pages = {685--693},\n  publisher = {American Association for the Advancement of Science},\n  issn = {0036-8075},\n  url = {https://www.jstor.org/stable/1740672},\n  urldate = {2024-03-20},\n  file = {/Users/shanest/sync/library/Gould/1975/Gould - 1975 - Honey Bee Recruitment The Dance-Language Controve.pdf}\n}\n\n
\n
\n\n\n\n
\n\n\n\n\n\n
\n
\n\n\n\n\n
\n\n\n \n\n \n \n \n \n\n
\n"}; document.write(bibbase_data.data);