Linguistic generalization and compositionality in modern artificial neural networks. Baroni, M. Philosophical Transactions of the Royal Society B: Biological Sciences, 2020. ![Linguistic generalization and compositionality in modern artificial neural networks [link]](https://bibbase.org/img/filetypes/link.svg "link")
Paper doi abstract bibtex 2 downloads In the last decade, deep artificial neural networks have achieved astounding performance in many natural language-processing tasks. Given the high productivity of language, these models must possess effective generalization abilities. It is widely assumed that humans handle linguistic productivity by means of algebraic compositional rules: are deep networks similarly compositional? After reviewing the main innovations characterizing current deep language-processing networks, I discuss a set of studies suggesting that deep networks are capable of subtle grammar-dependent generalizations, but also that they do not rely on systematic compositional rules. I argue that the intriguing behaviour of these devices (still awaiting a full understanding) should be of interest to linguists and cognitive scientists, as it offers a new perspective on possible computational strategies to deal with linguistic productivity beyond rule-based compositionality, and it might lead to new insights into the less systematic generalization patterns that also appear in natural language.
@article{Baroni2020,
abstract = {In the last decade, deep artificial neural networks have achieved astounding performance in many natural language-processing tasks. Given the high productivity of language, these models must possess effective generalization abilities. It is widely assumed that humans handle linguistic productivity by means of algebraic compositional rules: are deep networks similarly compositional? After reviewing the main innovations characterizing current deep language-processing networks, I discuss a set of studies suggesting that deep networks are capable of subtle grammar-dependent generalizations, but also that they do not rely on systematic compositional rules. I argue that the intriguing behaviour of these devices (still awaiting a full understanding) should be of interest to linguists and cognitive scientists, as it offers a new perspective on possible computational strategies to deal with linguistic productivity beyond rule-based compositionality, and it might lead to new insights into the less systematic generalization patterns that also appear in natural language.},
archivePrefix = {arXiv},
arxivId = {1904.00157},
author = {Baroni, Marco},
doi = {10.1098/rstb.2019.0307},
eprint = {1904.00157},
file = {:Users/shanest/Documents/Library/Baroni/Philosophical Transactions of the Royal Society B Biological Sciences/Baroni - 2020 - Linguistic generalization and compositionality in modern artificial neural networks.pdf:pdf},
issn = {0962-8436},
journal = {Philosophical Transactions of the Royal Society B: Biological Sciences},
keywords = {phenomenon: compositionality,survey},
number = {1791},
title = {{Linguistic generalization and compositionality in modern artificial neural networks}},
url = {https://royalsocietypublishing.org/doi/10.1098/rstb.2019.0307},
volume = {375},
year = {2020}
}
Downloads: 2
{"_id":"uBx8BXKXzrj7f7Rrq","bibbaseid":"baroni-linguisticgeneralizationandcompositionalityinmodernartificialneuralnetworks-2020","authorIDs":[],"author_short":["Baroni, M."],"bibdata":{"bibtype":"article","type":"article","abstract":"In the last decade, deep artificial neural networks have achieved astounding performance in many natural language-processing tasks. Given the high productivity of language, these models must possess effective generalization abilities. It is widely assumed that humans handle linguistic productivity by means of algebraic compositional rules: are deep networks similarly compositional? After reviewing the main innovations characterizing current deep language-processing networks, I discuss a set of studies suggesting that deep networks are capable of subtle grammar-dependent generalizations, but also that they do not rely on systematic compositional rules. I argue that the intriguing behaviour of these devices (still awaiting a full understanding) should be of interest to linguists and cognitive scientists, as it offers a new perspective on possible computational strategies to deal with linguistic productivity beyond rule-based compositionality, and it might lead to new insights into the less systematic generalization patterns that also appear in natural language.","archiveprefix":"arXiv","arxivid":"1904.00157","author":[{"propositions":[],"lastnames":["Baroni"],"firstnames":["Marco"],"suffixes":[]}],"doi":"10.1098/rstb.2019.0307","eprint":"1904.00157","file":":Users/shanest/Documents/Library/Baroni/Philosophical Transactions of the Royal Society B Biological Sciences/Baroni - 2020 - Linguistic generalization and compositionality in modern artificial neural networks.pdf:pdf","issn":"0962-8436","journal":"Philosophical Transactions of the Royal Society B: Biological Sciences","keywords":"phenomenon: compositionality,survey","number":"1791","title":"Linguistic generalization and compositionality in modern artificial neural networks","url":"https://royalsocietypublishing.org/doi/10.1098/rstb.2019.0307","volume":"375","year":"2020","bibtex":"@article{Baroni2020,\nabstract = {In the last decade, deep artificial neural networks have achieved astounding performance in many natural language-processing tasks. Given the high productivity of language, these models must possess effective generalization abilities. It is widely assumed that humans handle linguistic productivity by means of algebraic compositional rules: are deep networks similarly compositional? After reviewing the main innovations characterizing current deep language-processing networks, I discuss a set of studies suggesting that deep networks are capable of subtle grammar-dependent generalizations, but also that they do not rely on systematic compositional rules. I argue that the intriguing behaviour of these devices (still awaiting a full understanding) should be of interest to linguists and cognitive scientists, as it offers a new perspective on possible computational strategies to deal with linguistic productivity beyond rule-based compositionality, and it might lead to new insights into the less systematic generalization patterns that also appear in natural language.},\narchivePrefix = {arXiv},\narxivId = {1904.00157},\nauthor = {Baroni, Marco},\ndoi = {10.1098/rstb.2019.0307},\neprint = {1904.00157},\nfile = {:Users/shanest/Documents/Library/Baroni/Philosophical Transactions of the Royal Society B Biological Sciences/Baroni - 2020 - Linguistic generalization and compositionality in modern artificial neural networks.pdf:pdf},\nissn = {0962-8436},\njournal = {Philosophical Transactions of the Royal Society B: Biological Sciences},\nkeywords = {phenomenon: compositionality,survey},\nnumber = {1791},\ntitle = {{Linguistic generalization and compositionality in modern artificial neural networks}},\nurl = {https://royalsocietypublishing.org/doi/10.1098/rstb.2019.0307},\nvolume = {375},\nyear = {2020}\n}\n","author_short":["Baroni, M."],"key":"Baroni2020","id":"Baroni2020","bibbaseid":"baroni-linguisticgeneralizationandcompositionalityinmodernartificialneuralnetworks-2020","role":"author","urls":{"Paper":"https://royalsocietypublishing.org/doi/10.1098/rstb.2019.0307"},"keyword":["phenomenon: compositionality","survey"],"metadata":{"authorlinks":{}},"downloads":2},"bibtype":"article","biburl":"https://www.shane.st/teaching/575/win20/MachineLearning-interpretability.bib","creationDate":"2020-01-05T22:03:38.490Z","downloads":2,"keywords":["phenomenon: compositionality","survey"],"search_terms":["linguistic","generalization","compositionality","modern","artificial","neural","networks","baroni"],"title":"Linguistic generalization and compositionality in modern artificial neural networks","year":2020,"dataSources":["okYcdTpf4JJ2zkj7A","znj7izS5PeehdLR3G"]}