@inbook{
 type = {inbook},
 year = {2022},
 publisher = {CSLI},
 id = {51268105-189a-3b1c-add5-41127b77fc4b},
 created = {2022-06-08T16:20:02.556Z},
 file_attached = {true},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2022-06-15T16:20:24.963Z},
 read = {true},
 starred = {true},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Eshghi.etal22},
 private_publication = {false},
 bibtype = {inbook},
 author = {Eshghi, Arash and Gregoromichelaki, Eleni and Howes, Christine},
 editor = {Jean-Philippe Bernardy, undefined and Rasmus Blanck, undefined and Stergios Chatzikyriakidis, undefined and Shalom Lappin, undefined and Aleksandre Maskharashvili, undefined},
 chapter = {Action Coordination and Learning in Dialogue},
 title = {Probabilistic Approaches to Linguistic Theory}
}

@article{
 title = {Exploring Multi-Modal Representations for Ambiguity Detection & Coreference Resolution in the SIMMC 2.0 Challenge},
 type = {article},
 year = {2022},
 websites = {https://arxiv.org/abs/2202.12645,http://arxiv.org/abs/2202.12645},
 id = {b01712e2-d368-324d-acf6-c37e380076c9},
 created = {2022-06-08T17:11:32.014Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2022-06-08T17:58:11.364Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 source_type = {inproceedings},
 private_publication = {false},
 abstract = {Anaphoric expressions, such as pronouns and referential descriptions, are situated with respect to the linguistic context of prior turns, as well as, the immediate visual environment. However, a speaker's referential descriptions do not always uniquely identify the referent, leading to ambiguities in need of resolution through subsequent clarificational exchanges. Thus, effective Ambiguity Detection and Coreference Resolution are key to task success in Conversational AI. In this paper, we present models for these two tasks as part of the SIMMC 2.0 Challenge (Kottur et al. 2021). Specifically, we use TOD-BERT and LXMERT based models, compare them to a number of baselines and provide ablation experiments. Our results show that (1) language models are able to exploit correlations in the data to detect ambiguity; and (2) unimodal coreference resolution models can avoid the need for a vision component, through the use of smart object representations.},
 bibtype = {article},
 author = {Chiyah-Garcia, Francisco Javier and Suglia, Alessandro and Lopes, José and Eshghi, Arash and Hastie, Helen},
 journal = {Proceedings of AAAI 2022, DSTC10 Workshop}
}

Anaphoric expressions, such as pronouns and referential descriptions, are situated with respect to the linguistic context of prior turns, as well as, the immediate visual environment. However, a speaker's referential descriptions do not always uniquely identify the referent, leading to ambiguities in need of resolution through subsequent clarificational exchanges. Thus, effective Ambiguity Detection and Coreference Resolution are key to task success in Conversational AI. In this paper, we present models for these two tasks as part of the SIMMC 2.0 Challenge (Kottur et al. 2021). Specifically, we use TOD-BERT and LXMERT based models, compare them to a number of baselines and provide ablation experiments. Our results show that (1) language models are able to exploit correlations in the data to detect ambiguity; and (2) unimodal coreference resolution models can avoid the need for a vision component, through the use of smart object representations.

@inproceedings{
 title = {Combine to Describe: Evaluating Compositional Generalization in Image Captioning},
 type = {inproceedings},
 year = {2022},
 pages = {115-131},
 websites = {https://aclanthology.org/2022.acl-srw.11},
 month = {5},
 publisher = {Association for Computational Linguistics},
 city = {Dublin, Ireland},
 id = {8b180064-3282-3a97-9fdf-811e373a9dd5},
 created = {2022-06-08T18:13:28.563Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2022-06-08T18:13:28.563Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {pantazopoulos-etal-2022-combine},
 source_type = {inproceedings},
 private_publication = {false},
 abstract = {Compositionality -- the ability to combine simpler concepts to understand \& generate arbitrarily more complex conceptual structures -- has long been thought to be the cornerstone of human language capacity. With the recent, notable success of neural models in various NLP tasks, attention has now naturally turned to the compositional capacity of these models. In this paper, we study the compositional generalization properties of image captioning models. We perform a set experiments under controlled conditions using model and data ablations, each designed to benchmark a particular facet of compositional generalization: systematicity is the ability of a model to create novel combinations of concepts out of those observed during training, productivity is here operationalised as the capacity of a model to extend its predictions beyond the length distribution it has observed during training, and substitutivity is concerned with the robustness of the model against synonym substitutions. While previous work has focused primarily on systematicity, here we provide a more in-depth analysis of the strengths and weaknesses of state of the art captioning models. Our findings demonstrate that the models we study here do not compositionally generalize in terms of systematicity and productivity, however, they are robust to some degree to synonym substitutions},
 bibtype = {inproceedings},
 author = {Pantazopoulos, George and Suglia, Alessandro and Eshghi, Arash},
 doi = {10.18653/v1/2022.acl-srw.11},
 booktitle = {Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics: Student Research Workshop}
}

Compositionality -- the ability to combine simpler concepts to understand \& generate arbitrarily more complex conceptual structures -- has long been thought to be the cornerstone of human language capacity. With the recent, notable success of neural models in various NLP tasks, attention has now naturally turned to the compositional capacity of these models. In this paper, we study the compositional generalization properties of image captioning models. We perform a set experiments under controlled conditions using model and data ablations, each designed to benchmark a particular facet of compositional generalization: systematicity is the ability of a model to create novel combinations of concepts out of those observed during training, productivity is here operationalised as the capacity of a model to extend its predictions beyond the length distribution it has observed during training, and substitutivity is concerned with the robustness of the model against synonym substitutions. While previous work has focused primarily on systematicity, here we provide a more in-depth analysis of the strengths and weaknesses of state of the art captioning models. Our findings demonstrate that the models we study here do not compositionally generalize in terms of systematicity and productivity, however, they are robust to some degree to synonym substitutions

@article{
 title = {Feedback Relevance Spaces: Interactional Constraints on Processing Contexts in Dynamic Syntax},
 type = {article},
 year = {2021},
 pages = {331-362},
 volume = {30},
 websites = {https://doi.org/10.1007/s10849-020-09328-1},
 id = {5fb252bf-32e7-38e5-808e-7f1ac172c52a},
 created = {2022-06-08T16:15:25.990Z},
 file_attached = {true},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2022-06-08T17:58:10.883Z},
 read = {true},
 starred = {true},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 source_type = {JOUR},
 private_publication = {false},
 abstract = {Feedback such as backchannels and clarification requests often occurs subsententially, demonstrating the incremental nature of grounding in dialogue. However, although such feedback can occur at any point within an utterance, it typically does not do so, tending to occur at Feedback Relevance Spaces (FRSs). We present a corpus study of acknowledgements and clarification requests in British English, and describe how our low-level, semantic processing model in Dynamic Syntax accounts for this feedback. The model trivially accounts for the 85% of cases where feedback occurs at FRSs, but we also describe how it can be integrated or interpreted at non-FRSs using the predictive, incremental and interactive nature of the formalism. This model shows how feedback serves to continually realign processing contexts and thus manage the characteristic divergence and convergence that is key to moving dialogue forward.},
 bibtype = {article},
 author = {Howes, Christine and Eshghi, Arash},
 doi = {10.1007/s10849-020-09328-1},
 journal = {Journal of Logic, Language and Information},
 number = {2}
}

Feedback such as backchannels and clarification requests often occurs subsententially, demonstrating the incremental nature of grounding in dialogue. However, although such feedback can occur at any point within an utterance, it typically does not do so, tending to occur at Feedback Relevance Spaces (FRSs). We present a corpus study of acknowledgements and clarification requests in British English, and describe how our low-level, semantic processing model in Dynamic Syntax accounts for this feedback. The model trivially accounts for the 85% of cases where feedback occurs at FRSs, but we also describe how it can be integrated or interpreted at non-FRSs using the predictive, incremental and interactive nature of the formalism. This model shows how feedback serves to continually realign processing contexts and thus manage the characteristic divergence and convergence that is key to moving dialogue forward.

@inproceedings{
 title = {A study of automatic metrics for the evaluation of natural language explanations},
 type = {inproceedings},
 year = {2021},
 pages = {2376-2387},
 websites = {https://aclanthology.org/2021.eacl-main.202},
 month = {4},
 publisher = {Association for Computational Linguistics},
 city = {Online},
 id = {221688f6-e8b8-313e-9fcd-bcdb86740fce},
 created = {2022-06-08T17:11:31.944Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2022-06-08T17:58:11.393Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 source_type = {inproceedings},
 private_publication = {false},
 abstract = {As transparency becomes key for robotics and AI, it will be necessary to evaluate the methods through which transparency is provided, including automatically generated natural language (NL) explanations. Here, we explore parallels between the generation of such explanations and the much-studied field of evaluation of Natural Language Generation (NLG). Specifically, we investigate which of the NLG evaluation measures map well to explanations. We present the ExBAN corpus: a crowd-sourced corpus of NL explanations for Bayesian Networks. We run correlations comparing human subjective ratings with NLG automatic measures. We find that embedding-based automatic NLG evaluation methods, such as BERTScore and BLEURT, have a higher correlation with human ratings, compared to word-overlap metrics, such as BLEU and ROUGE. This work has implications for Explainable AI and transparent robotic and autonomous systems.},
 bibtype = {inproceedings},
 author = {Clinciu, Miruna Adriana and Eshghi, Arash and Hastie, Helen},
 doi = {10.18653/v1/2021.eacl-main.202},
 booktitle = {EACL 2021 - 16th Conference of the European Chapter of the Association for Computational Linguistics, Proceedings of the Conference}
}

As transparency becomes key for robotics and AI, it will be necessary to evaluate the methods through which transparency is provided, including automatically generated natural language (NL) explanations. Here, we explore parallels between the generation of such explanations and the much-studied field of evaluation of Natural Language Generation (NLG). Specifically, we investigate which of the NLG evaluation measures map well to explanations. We present the ExBAN corpus: a crowd-sourced corpus of NL explanations for Bayesian Networks. We run correlations comparing human subjective ratings with NLG automatic measures. We find that embedding-based automatic NLG evaluation methods, such as BERTScore and BLEURT, have a higher correlation with human ratings, compared to word-overlap metrics, such as BLEU and ROUGE. This work has implications for Explainable AI and transparent robotic and autonomous systems.

@article{
 title = {Incremental Graph-Based Semantics and Reasoning for Conversational AI},
 type = {article},
 year = {2021},
 pages = {1-7},
 websites = {https://aclanthology.org/2021.reinact-1.1},
 month = {10},
 publisher = {Association for Computational Linguistics},
 city = {Gothenburg, Sweden},
 id = {473049b8-380d-3261-a72d-d415a64587b9},
 created = {2022-06-08T17:11:32.155Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2022-06-08T17:58:11.363Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 source_type = {inproceedings},
 private_publication = {false},
 abstract = {The next generation of conversational AI systems need to: (1) process language incre-mentally, token-by-token to be more responsive and enable handling of conversational phenomena such as pauses, restarts and self-corrections; (2) reason incrementally allowing meaning to be established beyond what is said; (3) be transparent and controllable, allowing designers as well as the system itself to easily establish reasons for particular behaviour and tailor to particular user groups, or domains. In this short paper we present ongoing preliminary work combining Dynamic Syntax (DS)-an incremental, semantic grammar framework-with the Resource Description Framework (RDF). This paves the way for the creation of incremental semantic parsers that progressively output semantic RDF graphs as an utterance unfolds in real-time. We also outline how the parser can be integrated with an incremen-tal reasoning engine through RDF. We argue that this DS-RDF hybrid satisfies the desider-ata listed above, yielding semantic infrastructure that can be used to build responsive, real-time, interpretable Conversational AI that can be rapidly customised for specific user groups such as people with dementia.},
 bibtype = {article},
 author = {Addlesee, Angus and Eshghi, Arash},
 journal = {Proceedings of the Reasoning and Interaction Conference (ReInAct 2021)}
}

The next generation of conversational AI systems need to: (1) process language incre-mentally, token-by-token to be more responsive and enable handling of conversational phenomena such as pauses, restarts and self-corrections; (2) reason incrementally allowing meaning to be established beyond what is said; (3) be transparent and controllable, allowing designers as well as the system itself to easily establish reasons for particular behaviour and tailor to particular user groups, or domains. In this short paper we present ongoing preliminary work combining Dynamic Syntax (DS)-an incremental, semantic grammar framework-with the Resource Description Framework (RDF). This paves the way for the creation of incremental semantic parsers that progressively output semantic RDF graphs as an utterance unfolds in real-time. We also outline how the parser can be integrated with an incremen-tal reasoning engine through RDF. We argue that this DS-RDF hybrid satisfies the desider-ata listed above, yielding semantic infrastructure that can be used to build responsive, real-time, interpretable Conversational AI that can be rapidly customised for specific user groups such as people with dementia.

@inbook{
 type = {inbook},
 year = {2021},
 pages = {165-183},
 websites = {https://doi.org/10.1007/978-981-15-9323-9_15},
 publisher = {Springer Singapore},
 city = {Singapore},
 id = {d3a5ab1b-5cad-3e12-93b3-29cb49738d3b},
 created = {2022-06-08T17:58:10.412Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2022-06-08T17:58:10.412Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Liu.etal2021},
 source_type = {inbook},
 private_publication = {false},
 abstract = {We have recently seen the emergence of several publicly available Natural Language Understanding (NLU) toolkits, which map user utterances to structured, but more abstract, Dialogue Act (DA) or Intent specifications, while making this process accessible to the lay developer. In this paper, we present the first wide coverage evaluation and comparison of some of the most popular NLU services, on a large, multi-domain (21 domains) dataset of 25 K user utterances that we have collected and annotated with Intent and Entity Type specifications and which will be released as part of this submission (https://github.com/xliuhw/NLU-Evaluation-Data). The results show that on Intent classification Watson significantly outperforms the other platforms, namely, Dialogflow, LUIS and Rasa; though these also perform well. Interestingly, on Entity Type recognition, Watson performs significantly worse due to its low Precision (At the time of producing the camera-ready version of this paper, we noticed the seemingly recent addition of a `Contextual Entity' annotation tool to Watson, much like e.g. in Rasa. We'd threfore like to stress that this paper does not include an evaluation of this feature in Watson NLU.). Again, Dialogflow, LUIS and Rasa perform well on this task.},
 bibtype = {inbook},
 author = {Liu, Xingkun and Eshghi, Arash and Swietojanski, Pawel and Rieser, Verena},
 editor = {Marchi, Erik and Siniscalchi, Sabato Marco and Cumani, Sandro and Salerno, Valerio Mario and Li, Haizhou},
 doi = {10.1007/978-981-15-9323-9_15},
 chapter = {Benchmarking Natural Language Understanding Services for Building Conversational Agents},
 title = {Increasing Naturalness and Flexibility in Spoken Dialogue Interaction: 10th International Workshop on Spoken Dialogue Systems}
}

We have recently seen the emergence of several publicly available Natural Language Understanding (NLU) toolkits, which map user utterances to structured, but more abstract, Dialogue Act (DA) or Intent specifications, while making this process accessible to the lay developer. In this paper, we present the first wide coverage evaluation and comparison of some of the most popular NLU services, on a large, multi-domain (21 domains) dataset of 25 K user utterances that we have collected and annotated with Intent and Entity Type specifications and which will be released as part of this submission (https://github.com/xliuhw/NLU-Evaluation-Data). The results show that on Intent classification Watson significantly outperforms the other platforms, namely, Dialogflow, LUIS and Rasa; though these also perform well. Interestingly, on Entity Type recognition, Watson performs significantly worse due to its low Precision (At the time of producing the camera-ready version of this paper, we noticed the seemingly recent addition of a `Contextual Entity' annotation tool to Watson, much like e.g. in Rasa. We'd threfore like to stress that this paper does not include an evaluation of this feature in Watson NLU.). Again, Dialogflow, LUIS and Rasa perform well on this task.

@article{
 title = {Completability vs (In)completeness},
 type = {article},
 year = {2020},
 keywords = {Dynamic syntax,English,Modern Greek,ellipsis,fragments,incrementality,joint action,repair,split utterances},
 id = {cf0d5b6c-caf8-333c-8459-253f38e91700},
 created = {2020-10-31T23:59:00.000Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2024-01-15T20:03:34.740Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {© 2020 The Linguistic Circle of Copenhagen. In everyday conversation, no notion of “complete sentence” is required for syntactic licensing. However, so-called “fragmentary”, “incomplete”, and abandoned utterances are problematic for standard formalisms. When contextualised, such data show that (a) non-sentential utterances are adequate to underpin agent coordination, while (b) all linguistic dependencies can be systematically distributed across participants and turns. Standard models have problems accounting for such data because their notions of ‘constituency’ and ‘syntactic domain’ are independent of performance considerations. Concomitantly, we argue that no notion of “full proposition” or encoded speech act is necessary for successful interaction: strings, contents, and joint actions emerge in conversation without any single participant having envisaged in advance the outcome of their own or their interlocutors’ actions. Nonetheless, morphosyntactic and semantic licensing mechanisms need to apply incrementally and subsententially. We argue that, while a representational level of abstract syntax, divorced from conceptual structure and physical action, impedes natural accounts of subsentential coordination phenomena, a view of grammar as a “skill” employing domain-general mechanisms, rather than fixed form-meaning mappings, is needed instead. We provide a sketch of a predictive and incremental architecture (Dynamic Syntax) within which underspecification and time-relative update of meanings and utterances constitute the sole concept of “syntax”.},
 bibtype = {article},
 author = {Gregoromichelaki, E. and Mills, G.J. and Howes, C. and Eshghi, Arash and Chatzikyriakidis, S. and Purver, M. and Kempson, R. and Cann, R. and Healey, P.G.T.},
 doi = {10.1080/03740463.2020.1795549},
 journal = {Acta Linguistica Hafniensia}
}

© 2020 The Linguistic Circle of Copenhagen. In everyday conversation, no notion of “complete sentence” is required for syntactic licensing. However, so-called “fragmentary”, “incomplete”, and abandoned utterances are problematic for standard formalisms. When contextualised, such data show that (a) non-sentential utterances are adequate to underpin agent coordination, while (b) all linguistic dependencies can be systematically distributed across participants and turns. Standard models have problems accounting for such data because their notions of ‘constituency’ and ‘syntactic domain’ are independent of performance considerations. Concomitantly, we argue that no notion of “full proposition” or encoded speech act is necessary for successful interaction: strings, contents, and joint actions emerge in conversation without any single participant having envisaged in advance the outcome of their own or their interlocutors’ actions. Nonetheless, morphosyntactic and semantic licensing mechanisms need to apply incrementally and subsententially. We argue that, while a representational level of abstract syntax, divorced from conceptual structure and physical action, impedes natural accounts of subsentential coordination phenomena, a view of grammar as a “skill” employing domain-general mechanisms, rather than fixed form-meaning mappings, is needed instead. We provide a sketch of a predictive and incremental architecture (Dynamic Syntax) within which underspecification and time-relative update of meanings and utterances constitute the sole concept of “syntax”.

@inproceedings{
 title = {A Comprehensive Evaluation of Incremental Speech Recognition and Diarization for Conversational AI},
 type = {inproceedings},
 year = {2020},
 pages = {3492-3503},
 websites = {https://aclanthology.org/2020.coling-main.312},
 month = {12},
 publisher = {International Committee on Computational Linguistics},
 city = {Barcelona, Spain (Online)},
 id = {a1cfffbf-4759-335d-93f8-ff58d4435e4f},
 created = {2022-06-08T17:11:31.943Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2022-06-08T17:58:11.372Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 source_type = {inproceedings},
 private_publication = {false},
 abstract = {Automatic Speech Recognition (ASR) systems are increasingly powerful and more accurate, but also more numerous with several options existing currently as a service (e.g. Google, IBM, and Microsoft). Currently the most stringent standards for such systems are set within the context of their use in, and for, Conversational AI technology. These systems are expected to operate incrementally in real-time, be responsive, stable, and robust to the pervasive yet peculiar characteristics of conversational speech such as disfluencies and overlaps. In this paper we evaluate the most popular of such systems with metrics and experiments designed with these standards in mind. We also evaluate the speaker diarization (SD) capabilities of the same systems which will be particularly important for dialogue systems designed to handle multi-party interaction. We found that Microsoft has the leading incremental ASR system which preserves disfluent materials and IBM has the leading incremental SD system in addition to the ASR that is most robust to speech overlaps. Google strikes a balance between the two but none of these systems are yet suitable to reliably handle natural spontaneous conversations in real-time.},
 bibtype = {inproceedings},
 author = {Addlesee, Angus and Yu, Yanchao and Eshghi, Arash},
 doi = {10.18653/v1/2020.coling-main.312},
 booktitle = {Proceedings of the 28th International Conference on Computational Linguistics}
}

Automatic Speech Recognition (ASR) systems are increasingly powerful and more accurate, but also more numerous with several options existing currently as a service (e.g. Google, IBM, and Microsoft). Currently the most stringent standards for such systems are set within the context of their use in, and for, Conversational AI technology. These systems are expected to operate incrementally in real-time, be responsive, stable, and robust to the pervasive yet peculiar characteristics of conversational speech such as disfluencies and overlaps. In this paper we evaluate the most popular of such systems with metrics and experiments designed with these standards in mind. We also evaluate the speaker diarization (SD) capabilities of the same systems which will be particularly important for dialogue systems designed to handle multi-party interaction. We found that Microsoft has the leading incremental ASR system which preserves disfluent materials and IBM has the leading incremental SD system in addition to the ASR that is most robust to speech overlaps. Google strikes a balance between the two but none of these systems are yet suitable to reliably handle natural spontaneous conversations in real-time.

@inproceedings{
 title = {Affordance Competition in Dialogue: The Case of Syntactic Universals},
 type = {inproceedings},
 year = {2020},
 websites = {http://semdial.org/anthology/Z20-Gregoromichelaki_semdial_0022.pdf},
 month = {7},
 publisher = {SEMDIAL},
 city = {Virtually at Brandeis, Waltham, New Jersey},
 id = {b15fe848-6dee-35de-a2c8-c156efa61b1f},
 created = {2022-06-08T17:11:32.016Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2022-06-08T17:58:11.380Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 source_type = {inproceedings},
 private_publication = {false},
 bibtype = {inproceedings},
 author = {Gregoromichelaki, Eleni and Chatzikyriakidis, Stergios and Eshghi, Arash and Hough, Julian and Howes, Christine and Kempson, Ruth and Kiaer, Jieun and Purver, Matthew and Sadrzadeh, Mehrnoosh and White, Graham},
 booktitle = {Proceedings of the 24th Workshop on the Semantics and Pragmatics of Dialogue - Full Papers}
}

@inproceedings{
 title = {Benchmarking Natural Language Understanding Services for building Conversational Agents},
 type = {inproceedings},
 year = {2019},
 websites = {https://arxiv.org/abs/1903.05566},
 id = {57daac1b-fa3e-3de9-93bd-124af0f4ece7},
 created = {2019-04-10T11:14:23.351Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-04-10T11:14:23.351Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Liu.etal19},
 private_publication = {false},
 abstract = {We have recently seen the emergence of several publicly available Natural Language Understanding (NLU) toolkits, which map user utterances to structured, but more abstract, Dialogue Act (DA) or Intent specifications, while making this process accessible to the lay developer. In this paper, we present the first wide coverage evaluation and comparison of some of the most popular NLU services, on a large, multi-domain (21 domains) dataset of 25K user utterances that we have collected and annotated with Intent and Entity Type specifications and which will be released as part of this submission. The results show that on Intent classification Watson significantly outperforms the other platforms, namely, Dialogflow, LUIS and Rasa; though these also perform well. Interestingly, on Entity Type recognition, Watson performs significantly worse due to its low Precision. Again, Dialogflow, LUIS and Rasa perform well on this task.},
 bibtype = {inproceedings},
 author = {Liu, Xingkun and Eshghi, Arash and Swietojanski, Pawel and Rieser, Verena},
 booktitle = {Proceedings of the IWSDS 2019}
}

We have recently seen the emergence of several publicly available Natural Language Understanding (NLU) toolkits, which map user utterances to structured, but more abstract, Dialogue Act (DA) or Intent specifications, while making this process accessible to the lay developer. In this paper, we present the first wide coverage evaluation and comparison of some of the most popular NLU services, on a large, multi-domain (21 domains) dataset of 25K user utterances that we have collected and annotated with Intent and Entity Type specifications and which will be released as part of this submission. The results show that on Intent classification Watson significantly outperforms the other platforms, namely, Dialogflow, LUIS and Rasa; though these also perform well. Interestingly, on Entity Type recognition, Watson performs significantly worse due to its low Precision. Again, Dialogflow, LUIS and Rasa perform well on this task.

@inproceedings{
 title = {Interjection as coordination device: feedback relevance spaces},
 type = {inproceedings},
 year = {2019},
 websites = {https://drive.google.com/open?id=1uQNfET3_seCNc4z8GM7bcv3KOkuNGBzb},
 city = {Hong Kong},
 id = {693687e6-ac39-359e-9868-fc1efc531f0c},
 created = {2019-04-10T23:43:27.506Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-06-17T09:57:25.457Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Howes.Eshghi19},
 private_publication = {false},
 abstract = {Dialogue is co-constructed by multiple interlocutors with frequent feedback demonstrating whether something said is taken as understood [1,2]. To achieve this grounding, we produce relevant next turns or interjections (e.g. 'mm', 'yeah'). Some interjections indicate processing or coordination difficulties and the need for repair (e.g. 'huh?'). This feedback does not just occur at the ends of turns, but sub-sententially, showing that grounding occurs incrementally, before a complete proposition has been produced/processed [3,4]. However, despite evidence that speaker switch can occur anywhere, even within syntactic constituents [5], feedback is not appropriate just anywhere -- randomly placed backchannels disrupt the flow of dialogue, are rated as less natural and decrease rapport (e.g. [6]). Using Dynamic Syntax [7], we provide a low-level, semantic processing model of where feedback ought to be licensed -- feedback relevance spaces (FRSs). These are analogous to (but more common than) transition relevance places (TRPs; [8]) -- places where speaker switch may occur. Just as this is optional at TRPs, feedback is optional at FRSs. The model accounts for cases where feedback occurs at FRSs, and how it can be integrated at non-FRSs using the predictive, incremental and interactive nature of the formalism. In contrast to models of feedback that incorporate higher order reasoning about mental states (e.g. [9]), this model shows how feedback serves to continually realign processing contexts without recourse to higher order pragmatic reasoning, and provides a mechanistic model of the characteristic divergence and convergence that is key to moving dialogue forward. As well as providing insights into human-human communication, this work has implications for the production and interpretation of human-like feedback in dialogue systems; not just based on unanalysed features (which may result in accurate placement), but because they have successfully compiled a semantic unit. Our FRS model is implemented [10] and deployed in a new dialogue system architecture [11]; ongoing work explores the naturalness and usability of such systems versus, turn-based systems or those without the FRS model. [1] Bavelas, Coates, Johnson (2000). Listeners as co-narrators. Journal of personality and social psychology 79(6). [2] Clark (1996). Using Language. Cambridge University Press. [3] Eshghi, Howes, Hough, Gregoromichelaki, Purver (2015). Feedback in conversation as incremental semantic update. Proceedings IWCS. [4] Howes, Eshghi (2017). Feedback relevance spaces: The organisation of increments in conversation. Proceedings IWCS. [5] Howes, Purver, Healey, Mills, Gregoromichelaki (2011). On incrementality in dialogue: Evidence from compound contributions. Dialogue and Discourse 2(1). [6] Poppe, Truong, Heylen (2011). Backchannels: Quantity, type and timing matters. International Workshop on Intelligent Virtual Agents. [7] Kempson, Cann, Gregoromichelaki, Chatzikiriakidis (2016). Language as mechanisms for interaction. Theoretical Linguistics 42(3-4). [8] Sacks, Schegloff, Jefferson (1974). A simplest systematics for the organization of turntaking for conversation. Language 50(4). [9] Visser, Traum, DeVault, op den Akker (2014). A model for incremental grounding in spoken dialogue systems. Journal on Multimodal User Interfaces 8(1). [10] Eshghi (2015). Dylan: An incremental, semantic, contextual parser for dialogue. Proceedings SEMDIAL. [11] Eshghi, Shalyminov, Lemon. (2017) Bootstrapping incremental dialogue systems from minimal data: linguistic knowledge or machine learning? Proceedings EMNLP.},
 bibtype = {inproceedings},
 author = {Howes, Christine and Eshghi, Arash},
 booktitle = {Proceedings of the 16th International Pragmatics Conference}
}

Dialogue is co-constructed by multiple interlocutors with frequent feedback demonstrating whether something said is taken as understood [1,2]. To achieve this grounding, we produce relevant next turns or interjections (e.g. 'mm', 'yeah'). Some interjections indicate processing or coordination difficulties and the need for repair (e.g. 'huh?'). This feedback does not just occur at the ends of turns, but sub-sententially, showing that grounding occurs incrementally, before a complete proposition has been produced/processed [3,4]. However, despite evidence that speaker switch can occur anywhere, even within syntactic constituents [5], feedback is not appropriate just anywhere -- randomly placed backchannels disrupt the flow of dialogue, are rated as less natural and decrease rapport (e.g. [6]). Using Dynamic Syntax [7], we provide a low-level, semantic processing model of where feedback ought to be licensed -- feedback relevance spaces (FRSs). These are analogous to (but more common than) transition relevance places (TRPs; [8]) -- places where speaker switch may occur. Just as this is optional at TRPs, feedback is optional at FRSs. The model accounts for cases where feedback occurs at FRSs, and how it can be integrated at non-FRSs using the predictive, incremental and interactive nature of the formalism. In contrast to models of feedback that incorporate higher order reasoning about mental states (e.g. [9]), this model shows how feedback serves to continually realign processing contexts without recourse to higher order pragmatic reasoning, and provides a mechanistic model of the characteristic divergence and convergence that is key to moving dialogue forward. As well as providing insights into human-human communication, this work has implications for the production and interpretation of human-like feedback in dialogue systems; not just based on unanalysed features (which may result in accurate placement), but because they have successfully compiled a semantic unit. Our FRS model is implemented [10] and deployed in a new dialogue system architecture [11]; ongoing work explores the naturalness and usability of such systems versus, turn-based systems or those without the FRS model. [1] Bavelas, Coates, Johnson (2000). Listeners as co-narrators. Journal of personality and social psychology 79(6). [2] Clark (1996). Using Language. Cambridge University Press. [3] Eshghi, Howes, Hough, Gregoromichelaki, Purver (2015). Feedback in conversation as incremental semantic update. Proceedings IWCS. [4] Howes, Eshghi (2017). Feedback relevance spaces: The organisation of increments in conversation. Proceedings IWCS. [5] Howes, Purver, Healey, Mills, Gregoromichelaki (2011). On incrementality in dialogue: Evidence from compound contributions. Dialogue and Discourse 2(1). [6] Poppe, Truong, Heylen (2011). Backchannels: Quantity, type and timing matters. International Workshop on Intelligent Virtual Agents. [7] Kempson, Cann, Gregoromichelaki, Chatzikiriakidis (2016). Language as mechanisms for interaction. Theoretical Linguistics 42(3-4). [8] Sacks, Schegloff, Jefferson (1974). A simplest systematics for the organization of turntaking for conversation. Language 50(4). [9] Visser, Traum, DeVault, op den Akker (2014). A model for incremental grounding in spoken dialogue systems. Journal on Multimodal User Interfaces 8(1). [10] Eshghi (2015). Dylan: An incremental, semantic, contextual parser for dialogue. Proceedings SEMDIAL. [11] Eshghi, Shalyminov, Lemon. (2017) Bootstrapping incremental dialogue systems from minimal data: linguistic knowledge or machine learning? Proceedings EMNLP.

@inproceedings{
 title = {Current Challenges in Spoken Dialogue Systems and Why they are Critical for Healthcare Applications},
 type = {inproceedings},
 year = {2019},
 city = {Stockholm, Sweden},
 id = {0221973b-2f4b-3fd5-845a-89f321fd5676},
 created = {2019-08-16T08:41:19.724Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-08-16T08:41:19.724Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Addlesee.etal19},
 private_publication = {false},
 abstract = {Dialogue technology such as Amazon's Alexa has the potential to transform the healthcare industry. However, current systems are not yet naturally interactive: they are often turn-based, have naive end-of-turn detection and completely ignore many types of verbal and visual feedback - such as backchannels, hesitation markers, filled pauses, gaze, brow furrows & disfluencies - that are crucial in guiding & managing the conversational process. This is especially important in the healthcare industry as target users of Spoken Dialogue Systems (SDS) are likely to be frail, older, distracted, or suffer from cognitive decline, which impact their ability to make effective use of current systems. In this paper, we outline some of the challenges that are in urgent need of further research, including Incremental Speech Recognition, Natural Language Understanding and a systematic study of the interactional patterns in conversation that are potentially diagnostic of Dementia, and how these might inform research on and the design of the next generation of SDSs.},
 bibtype = {inproceedings},
 author = {Addlesee, Angus and Konstas, Ioannis and Eshghi, Arash},
 booktitle = {Proceedings of the Dialogue for Good (DiGo) 2019 workshop}
}

Dialogue technology such as Amazon's Alexa has the potential to transform the healthcare industry. However, current systems are not yet naturally interactive: they are often turn-based, have naive end-of-turn detection and completely ignore many types of verbal and visual feedback - such as backchannels, hesitation markers, filled pauses, gaze, brow furrows & disfluencies - that are crucial in guiding & managing the conversational process. This is especially important in the healthcare industry as target users of Spoken Dialogue Systems (SDS) are likely to be frail, older, distracted, or suffer from cognitive decline, which impact their ability to make effective use of current systems. In this paper, we outline some of the challenges that are in urgent need of further research, including Incremental Speech Recognition, Natural Language Understanding and a systematic study of the interactional patterns in conversation that are potentially diagnostic of Dementia, and how these might inform research on and the design of the next generation of SDSs.

@inproceedings{
 title = {Few-Shot Dialogue Generation Without Annotated Data: A Transfer Learning Approach},
 type = {inproceedings},
 year = {2019},
 websites = {https://arxiv.org/abs/1908.05854},
 publisher = {ACL},
 city = {Stockholm, Sweden},
 id = {99f33106-6acf-34d0-9a9a-371aa2cd305b},
 created = {2019-08-16T08:41:19.815Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-08-20T09:50:49.213Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Shalyminov.etal19b},
 private_publication = {false},
 abstract = {Learning with minimal data is one of the key challenges in the development of practical, production-ready goal-oriented dialogue systems. In a real-world enterprise setting where dialogue systems are developed rapidly and are expected to work robustly for an ever-growing variety of domains, products, and scenarios, efficient learning from a limited number of examples becomes indispensable. In this paper, we introduce a technique to achieve state-of-the-art dialogue generation performance in a few-shot setup, without using any annotated data. We do this by leveraging background knowledge from a larger, more highly represented dialogue source --- namely, the MetaLWOz dataset. We evaluate our model on the Stanford Multi-Domain Dialogue Dataset, consisting of human-human goal-oriented dialogues in in-car navigation, appointment scheduling, and weather information domains. We show that our few-shot approach achieves state-of-the art results on that dataset by consistently outperforming the previous best model in terms of BLEU and Entity F1 scores, while being more data-efficient by not requiring any data annotation.},
 bibtype = {inproceedings},
 author = {Shalyminov, Igor and Lemon, Oliver and Eshghi, Arash and Lee, Sungjin},
 booktitle = {Proceedings of SIGdial 2019},
 keywords = {Dialogue System,Few-Shot,Generation,Knowledge Transfer,Transfer Learning}
}

Learning with minimal data is one of the key challenges in the development of practical, production-ready goal-oriented dialogue systems. In a real-world enterprise setting where dialogue systems are developed rapidly and are expected to work robustly for an ever-growing variety of domains, products, and scenarios, efficient learning from a limited number of examples becomes indispensable. In this paper, we introduce a technique to achieve state-of-the-art dialogue generation performance in a few-shot setup, without using any annotated data. We do this by leveraging background knowledge from a larger, more highly represented dialogue source --- namely, the MetaLWOz dataset. We evaluate our model on the Stanford Multi-Domain Dialogue Dataset, consisting of human-human goal-oriented dialogues in in-car navigation, appointment scheduling, and weather information domains. We show that our few-shot approach achieves state-of-the art results on that dataset by consistently outperforming the previous best model in terms of BLEU and Entity F1 scores, while being more data-efficient by not requiring any data annotation.

@inproceedings{
 title = {Normativity, Meaning Plasticity, and the Significance of Vector Space Semantics},
 type = {inproceedings},
 year = {2019},
 id = {733399ff-8b17-3bd8-9fa0-4b07c8d24a6a},
 created = {2019-08-16T08:41:19.847Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-08-16T08:41:19.847Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Gregoromichelaki19},
 private_publication = {false},
 abstract = {This paper continues the discussion started in Lucking et. al (2019), on the suitability or otherwise of Vector Space Semantics (VSS) as a model of semantics for NL in interaction.},
 bibtype = {inproceedings},
 author = {Gregoromichelaki, Eleni and Howes, Christine and Eshghi, Arash and Kempson, Ruth and Sadrzadeh, Mehrnoosh and Hough, Julian and Purver, Matthew and Wijnholds, Gijs},
 editor = {Howes, Christine and Hough, Julian},
 booktitle = {Proceedings of SemDial 2019 (LondonLogue)}
}

This paper continues the discussion started in Lucking et. al (2019), on the suitability or otherwise of Vector Space Semantics (VSS) as a model of semantics for NL in interaction.

@inproceedings{
 title = {“What are you laughing at?” Incremental processing of laughter in interaction},
 type = {inproceedings},
 year = {2019},
 city = {London},
 id = {0b662523-908a-31e3-899e-fd3ebb7d079a},
 created = {2019-08-16T08:41:19.930Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-08-20T09:53:51.750Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Eshghi.etal19},
 private_publication = {false},
 abstract = {In dialogue, laughter is frequent and can precede, follow or overlap what is laughed at. In this paper, we provide a prelimary, but unitary formal account of how forward- & backward- looking laughter are processed and integrated within the Dynamic Syntax framework.},
 bibtype = {inproceedings},
 author = {Eshghi, Arash and Maraev, Vladislav and Howes, Christine and Hough, Julian and Mazzocconi, Chiara},
 editor = {Howes, Christine and Hough, Julian},
 booktitle = {Proceedings of SemDial 2019 (LondonLogue)}
}

In dialogue, laughter is frequent and can precede, follow or overlap what is laughed at. In this paper, we provide a prelimary, but unitary formal account of how forward- & backward- looking laughter are processed and integrated within the Dynamic Syntax framework.

@inproceedings{
 title = {Data-Efficient Goal-Oriented Conversation with Dialogue Knowledge Transfer Networks},
 type = {inproceedings},
 year = {2019},
 publisher = {ACL},
 id = {67b14bd2-e20f-3534-a1e5-6b204b24ca43},
 created = {2019-08-16T08:41:20.021Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-08-16T08:41:20.021Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Shalyminov.etal19},
 private_publication = {false},
 bibtype = {inproceedings},
 author = {Shalyminov, Igor and Lee, Sungjin and Eshghi, Arash and Lemon, Oliver},
 booktitle = {Proceedings of EMNLP-IJCNLP 2019}
}

@misc{
 title = {Proceedings of the IWCS Workshop: Vector Semantics for Dialogue and Discourse},
 type = {misc},
 year = {2019},
 websites = {https://sites.google.com/site/dialoguevector/home},
 id = {4726d637-86ce-3a43-8399-0c5fc89547a6},
 created = {2019-08-16T08:51:10.712Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-08-20T09:42:18.566Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Sadrzadeh.etal19},
 private_publication = {false},
 bibtype = {misc},
 author = {Sadrzadeh, Mehrnoosh and Purver, Matthew and Eshghi, Arash and Hough, Julian and Kempson, Ruth and Healey, Patrick G.T.},
 editor = {Sadrzadeh, Mehrnoosh and Purver, Matthew and Eshghi, Arash and Hough, Julian and Kempson, Ruth and Healey, Patrick G.T.}
}

@misc{
 title = {Benchmarking natural language understanding services for building conversational agents},
 type = {misc},
 year = {2019},
 source = {arXiv},
 id = {3322a4de-833a-3fd0-a6ad-ad82e005daf7},
 created = {2020-10-27T23:59:00.000Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2020-10-30T08:03:31.996Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {Copyright © 2019, arXiv, All rights reserved. We have recently seen the emergence of several publicly available Natural Language Understanding (NLU) toolkits, which map user utterances to structured, but more abstract, Dialogue Act (DA) or Intent specifications, while making this process accessible to the lay developer. In this paper, we present the first wide coverage evaluation and comparison of some of the most popular NLU services, on a large, multi-domain (21 domains) dataset of 25K user utterances that we have collected and annotated with Intent and Entity Type specifications and which will be released as part of this submission.1 The results show that on Intent classification Watson significantly outperforms the other platforms, namely, Dialogflow, LUIS and Rasa; though these also perform well. Interestingly, on Entity Type recognition, Watson performs significantly worse due to its low Precision2. Again, Dialogflow, LUIS and Rasa perform well on this task.},
 bibtype = {misc},
 author = {Liu, X. and Eshghi, A. and Swietojanski, P. and Rieser, V.}
}

Copyright © 2019, arXiv, All rights reserved. We have recently seen the emergence of several publicly available Natural Language Understanding (NLU) toolkits, which map user utterances to structured, but more abstract, Dialogue Act (DA) or Intent specifications, while making this process accessible to the lay developer. In this paper, we present the first wide coverage evaluation and comparison of some of the most popular NLU services, on a large, multi-domain (21 domains) dataset of 25K user utterances that we have collected and annotated with Intent and Entity Type specifications and which will be released as part of this submission.1 The results show that on Intent classification Watson significantly outperforms the other platforms, namely, Dialogflow, LUIS and Rasa; though these also perform well. Interestingly, on Entity Type recognition, Watson performs significantly worse due to its low Precision2. Again, Dialogflow, LUIS and Rasa perform well on this task.

@misc{
 title = {Data-efficient goal-oriented conversation with dialogue knowledge transfer networks},
 type = {misc},
 year = {2019},
 source = {arXiv},
 id = {aa4adda5-5691-30ec-a0ae-17c626fef900},
 created = {2020-11-06T23:59:00.000Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2020-11-10T23:36:50.345Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {Copyright © 2019, arXiv, All rights reserved. Goal-oriented dialogue systems are now being widely adopted in industry where it is of key importance to maintain a rapid prototyping cycle for new products and domains. Data-driven dialogue system development has to be adapted to meet this requirement — therefore, reducing the amount of data and annotations necessary for training such systems is a central research problem. In this paper, we present the Dialogue Knowledge Transfer Network (DiKTNet), a state-of-the-art approach to goal-oriented dialogue generation which only uses a few example dialogues (i.e. few-shot learning), none of which has to be annotated. We achieve this by performing a 2-stage training. Firstly, we perform unsupervised dialogue representation pre-training on a large source of goal-oriented dialogues in multiple domains, the MetaLWOz corpus. Secondly, at the transfer stage, we train DiKTNet using this representation together with 2 other textual knowledge sources with different levels of generality: ELMo encoder and the main dataset’s source domains. Our main dataset is the Stanford Multi-Domain dialogue corpus. We evaluate our model on it in terms of BLEU and Entity F1 scores, and show that our approach significantly and consistently improves upon a series of baseline models as well as over the previous state-of-the-art dialogue generation model, ZSDG. The improvement upon the latter — up to 10% in Entity F1 and the average of 3% in BLEU score — is achieved using only the equivalent of 10% of ZSDG’s in-domain training data.},
 bibtype = {misc},
 author = {Shalyminov, I. and Lee, S. and Eshghi, A. and Lemon, O.}
}

Copyright © 2019, arXiv, All rights reserved. Goal-oriented dialogue systems are now being widely adopted in industry where it is of key importance to maintain a rapid prototyping cycle for new products and domains. Data-driven dialogue system development has to be adapted to meet this requirement — therefore, reducing the amount of data and annotations necessary for training such systems is a central research problem. In this paper, we present the Dialogue Knowledge Transfer Network (DiKTNet), a state-of-the-art approach to goal-oriented dialogue generation which only uses a few example dialogues (i.e. few-shot learning), none of which has to be annotated. We achieve this by performing a 2-stage training. Firstly, we perform unsupervised dialogue representation pre-training on a large source of goal-oriented dialogues in multiple domains, the MetaLWOz corpus. Secondly, at the transfer stage, we train DiKTNet using this representation together with 2 other textual knowledge sources with different levels of generality: ELMo encoder and the main dataset’s source domains. Our main dataset is the Stanford Multi-Domain dialogue corpus. We evaluate our model on it in terms of BLEU and Entity F1 scores, and show that our approach significantly and consistently improves upon a series of baseline models as well as over the previous state-of-the-art dialogue generation model, ZSDG. The improvement upon the latter — up to 10% in Entity F1 and the average of 3% in BLEU score — is achieved using only the equivalent of 10% of ZSDG’s in-domain training data.

@inproceedings{
 title = {``What are you laughing at?'' Incremental Processing of Laughter in Interaction},
 type = {inproceedings},
 year = {2019},
 websites = {http://semdial.org/anthology/Z19-Eshghi_semdial_0034.pdf},
 month = {9},
 publisher = {SEMDIAL},
 city = {London, United Kingdom},
 id = {c91ffeaa-e13e-3f2e-b4b2-40339396b19f},
 created = {2022-06-08T17:11:31.793Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2022-06-08T17:58:11.388Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 source_type = {inproceedings},
 private_publication = {false},
 bibtype = {inproceedings},
 author = {Eshghi, Arash and Maraev, Vladislav and Howes, Christine and Hough, Julian and Mazzocconi, Chiara},
 booktitle = {Proceedings of the 23rd Workshop on the Semantics and Pragmatics of Dialogue - Poster Abstracts}
}

@inproceedings{
 title = {Alana v2: Entertaining and Informative Open-domain Social Dialogue using Ontologies and Entity Linking},
 type = {inproceedings},
 year = {2018},
 websites = {https://s3.amazonaws.com/dex-microsites-prod/alexaprize/2018/papers/Alana.pdf},
 publisher = {Amazon},
 id = {72421b09-b4c9-30a2-aab9-1ed4bc449b24},
 created = {2019-04-16T14:28:35.519Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-06-16T14:48:19.295Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Curry.etal18},
 private_publication = {false},
 abstract = {We describe our 2018 Alexa prize system (called ‘Alana’) which consists of an ensemble of bots, combining rule-based and machine learning systems. This paper reports on the version of the system developed and evaluated in the semifinals of the 2018 competition (i.e. up to 15 August 2018), but not on subsequent enhancements. The main advances over our 2017 Alana system are: (1) a deeper Natural Language Understanding (NLU) pipeline; (2) the use of topic ontologies and Named Entity Linking to enable the user to navigate and search through a web of related information; rendering Alana in part an interactive NL interface to linked information on the web; (3) system generated clarification questions to interactively disambiguate between Named Entities as part of NLU; (4) a new profanity & abuse detection model with rule-based mitigation strategies; and (5) response retrieval from Reddit. We also present several ablation studies that measure the performance contributions of specific features (e.g. use of Ontology-bot, Reddit-bot, rule-based systems, etc). We find that these features increase overall system performance. Our final score, namely averaged user ratings over the whole semi-finals period, was 3.4. We were also able to achieve long dialogues (average around 11 turns and 2.20 minutes) during the semi-finals period.},
 bibtype = {inproceedings},
 author = {Amanda Cercas Curry, undefined and Papaioannou, Ioannis and Suglia, Alessandro and Agarwal, Shubham and Shalyminov, Igor and Xu, Xinnuo and Dušek, Ondřej and Eshghi, Arash and Konstas, Ioannis and Rieser, Verena and Lemon, Oliver},
 booktitle = {Alexa Prize Proceedings, Amazon RE-INVENT}
}

We describe our 2018 Alexa prize system (called ‘Alana’) which consists of an ensemble of bots, combining rule-based and machine learning systems. This paper reports on the version of the system developed and evaluated in the semifinals of the 2018 competition (i.e. up to 15 August 2018), but not on subsequent enhancements. The main advances over our 2017 Alana system are: (1) a deeper Natural Language Understanding (NLU) pipeline; (2) the use of topic ontologies and Named Entity Linking to enable the user to navigate and search through a web of related information; rendering Alana in part an interactive NL interface to linked information on the web; (3) system generated clarification questions to interactively disambiguate between Named Entities as part of NLU; (4) a new profanity & abuse detection model with rule-based mitigation strategies; and (5) response retrieval from Reddit. We also present several ablation studies that measure the performance contributions of specific features (e.g. use of Ontology-bot, Reddit-bot, rule-based systems, etc). We find that these features increase overall system performance. Our final score, namely averaged user ratings over the whole semi-finals period, was 3.4. We were also able to achieve long dialogues (average around 11 turns and 2.20 minutes) during the semi-finals period.

@article{
 title = {Running Repairs: Coordinating Meaning in Dialogue},
 type = {article},
 year = {2018},
 keywords = {Dialogue,Miscommunication,Repair},
 pages = {367-388},
 volume = {10},
 id = {6025321b-4ed7-3e3e-aff7-b65ace6b8fa4},
 created = {2019-08-20T09:40:40.145Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2022-06-08T17:58:10.857Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 private_publication = {false},
 abstract = {People give feedback in conversation: both positive signals of understanding, such as nods, and negative signals of misunderstanding, such as frowns. How do signals of understanding and misunderstanding affect the coordination of language use in conversation? Using a chat tool and a maze-based reference task, we test two experimental manipulations that selectively interfere with feedback in live conversation: (a) “Attenuation” that replaces positive signals of understanding such as “right” or “okay” with weaker, more provisional signals such as “errr” or “umm” and (2) “Amplification” that replaces relatively specific signals of misunderstanding from clarification requests such as “on the left?” with generic signals of trouble such as “huh?” or “eh?”. The results show that Amplification promotes rapid convergence on more systematic, abstract ways of describing maze locations while Attenuation has no significant effect. We interpret this as evidence that “running repairs”—the processes of dealing with misunderstandings on the fly—are key drivers of semantic coordination in dialogue. This suggests a new direction for experimental work on conversation and a productive way to connect the empirical accounts of Conversation Analysis with the representational and processing concerns of Formal Semantics and Psycholinguistics.},
 bibtype = {article},
 author = {Healey, Patrick G.T. and Mills, Gregory J. and Eshghi, Arash and Howes, Christine},
 doi = {10.1111/tops.12336},
 journal = {Topics in Cognitive Science},
 number = {2}
}

People give feedback in conversation: both positive signals of understanding, such as nods, and negative signals of misunderstanding, such as frowns. How do signals of understanding and misunderstanding affect the coordination of language use in conversation? Using a chat tool and a maze-based reference task, we test two experimental manipulations that selectively interfere with feedback in live conversation: (a) “Attenuation” that replaces positive signals of understanding such as “right” or “okay” with weaker, more provisional signals such as “errr” or “umm” and (2) “Amplification” that replaces relatively specific signals of misunderstanding from clarification requests such as “on the left?” with generic signals of trouble such as “huh?” or “eh?”. The results show that Amplification promotes rapid convergence on more systematic, abstract ways of describing maze locations while Attenuation has no significant effect. We interpret this as evidence that “running repairs”—the processes of dealing with misunderstandings on the fly—are key drivers of semantic coordination in dialogue. This suggests a new direction for experimental work on conversation and a productive way to connect the empirical accounts of Conversation Analysis with the representational and processing concerns of Formal Semantics and Psycholinguistics.

@inproceedings{
 title = {Challenging Neural Dialogue Models with Natural Data: Memory Networks Fail on Incremental Phenomena},
 type = {inproceedings},
 year = {2017},
 websites = {https://arxiv.org/pdf/1709.07840.pdf},
 city = {Barcelona},
 id = {027edae7-4406-320b-bd2b-f826a1234fee},
 created = {2019-04-10T10:35:05.562Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-04-10T10:35:05.562Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Shalyminov.etal17},
 source_type = {inproceedings},
 private_publication = {false},
 bibtype = {inproceedings},
 author = {Shalyminov, Igor and Eshghi, Arash and Lemon, Oliver},
 booktitle = {Proceedings of the 21st Workshop on the Semantics and Pragmatics of Dialogue (SemDial 2017 - SaarDial)}
}

@inbook{
 type = {inbook},
 year = {2017},
 websites = {https://drive.google.com/open?id=181YOCH5opX-4FEhymHXFd52z-zfDr11w},
 publisher = {Oxford University Press},
 id = {59911dc6-efdf-3b45-8044-a20cf17845f6},
 created = {2019-04-10T10:35:05.570Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-06-16T14:48:19.276Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Kempson.etal17},
 source_type = {incollection},
 private_publication = {false},
 bibtype = {inbook},
 author = {Kempson, Ruth and Gregoromichelaki, Eleni and Eshghi, Arash and Hough, Julian},
 chapter = {Ellipsis in Dynamic Syntax},
 title = {Oxford Handbook of Ellipsis}
}

@article{
 title = {Grammars as Mechanisms for Interaction: The Emergence of Language Games},
 type = {article},
 year = {2017},
 pages = {129-133},
 volume = {43},
 websites = {https://drive.google.com/open?id=1A2QFaES2SfVgKW5O_r7hL-Aq4z3t1vyx},
 id = {806d7dff-ce84-38eb-9510-69603bc871b3},
 created = {2019-04-10T10:35:05.572Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-06-17T09:57:25.482Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Eshghi.Lemon17},
 source_type = {article},
 private_publication = {false},
 abstract = {Dialogue is co-constructed by multiple interlocutors with frequent feedback demonstrating whether something said is taken as understood [1,2]. To achieve this grounding, we produce relevant next turns or interjections (e.g. 'mm', 'yeah'). Some interjections indicate processing or coordination difficulties and the need for repair (e.g. 'huh?'). This feedback does not just occur at the ends of turns, but sub-sententially, showing that grounding occurs incrementally, before a complete proposition has been produced/processed [3,4]. However, despite evidence that speaker switch can occur anywhere, even within syntactic constituents [5], feedback is not appropriate just anywhere -- randomly placed backchannels disrupt the flow of dialogue, are rated as less natural and decrease rapport (e.g. [6]). Using Dynamic Syntax [7], we provide a low-level, semantic processing model of where feedback ought to be licensed -- feedback relevance spaces (FRSs). These are analogous to (but more common than) transition relevance places (TRPs; [8]) -- places where speaker switch may occur. Just as this is optional at TRPs, feedback is optional at FRSs. The model accounts for cases where feedback occurs at FRSs, and how it can be integrated at non-FRSs using the predictive, incremental and interactive nature of the formalism. In contrast to models of feedback that incorporate higher order reasoning about mental states (e.g. [9]), this model shows how feedback serves to continually realign processing contexts without recourse to higher order pragmatic reasoning, and provides a mechanistic model of the characteristic divergence and convergence that is key to moving dialogue forward. As well as providing insights into human-human communication, this work has implications for the production and interpretation of human-like feedback in dialogue systems; not just based on unanalysed features (which may result in accurate placement), but because they have successfully compiled a semantic unit. Our FRS model is implemented [10] and deployed in a new dialogue system architecture [11]; ongoing work explores the naturalness and usability of such systems versus, turn-based systems or those without the FRS model. [1] Bavelas, Coates, Johnson (2000). Listeners as co-narrators. Journal of personality and social psychology 79(6). [2] Clark (1996). Using Language. Cambridge University Press. [3] Eshghi, Howes, Hough, Gregoromichelaki, Purver (2015). Feedback in conversation as incremental semantic update. Proceedings IWCS. [4] Howes, Eshghi (2017). Feedback relevance spaces: The organisation of increments in conversation. Proceedings IWCS. [5] Howes, Purver, Healey, Mills, Gregoromichelaki (2011). On incrementality in dialogue: Evidence from compound contributions. Dialogue and Discourse 2(1). [6] Poppe, Truong, Heylen (2011). Backchannels: Quantity, type and timing matters. International Workshop on Intelligent Virtual Agents. [7] Kempson, Cann, Gregoromichelaki, Chatzikiriakidis (2016). Language as mechanisms for interaction. Theoretical Linguistics 42(3-4). [8] Sacks, Schegloff, Jefferson (1974). A simplest systematics for the organization of turntaking for conversation. Language 50(4). [9] Visser, Traum, DeVault, op den Akker (2014). A model for incremental grounding in spoken dialogue systems. Journal on Multimodal User Interfaces 8(1). [10] Eshghi (2015). Dylan: An incremental, semantic, contextual parser for dialogue. Proceedings SEMDIAL. [11] Eshghi, Shalyminov, Lemon. (2017) Bootstrapping incremental dialogue systems from minimal data: linguistic knowledge or machine learning? Proceedings EMNLP.},
 bibtype = {article},
 author = {Eshghi, Arash and Lemon, Oliver},
 journal = {Theoretical Linguistics},
 number = {1-2}
}

Dialogue is co-constructed by multiple interlocutors with frequent feedback demonstrating whether something said is taken as understood [1,2]. To achieve this grounding, we produce relevant next turns or interjections (e.g. 'mm', 'yeah'). Some interjections indicate processing or coordination difficulties and the need for repair (e.g. 'huh?'). This feedback does not just occur at the ends of turns, but sub-sententially, showing that grounding occurs incrementally, before a complete proposition has been produced/processed [3,4]. However, despite evidence that speaker switch can occur anywhere, even within syntactic constituents [5], feedback is not appropriate just anywhere -- randomly placed backchannels disrupt the flow of dialogue, are rated as less natural and decrease rapport (e.g. [6]). Using Dynamic Syntax [7], we provide a low-level, semantic processing model of where feedback ought to be licensed -- feedback relevance spaces (FRSs). These are analogous to (but more common than) transition relevance places (TRPs; [8]) -- places where speaker switch may occur. Just as this is optional at TRPs, feedback is optional at FRSs. The model accounts for cases where feedback occurs at FRSs, and how it can be integrated at non-FRSs using the predictive, incremental and interactive nature of the formalism. In contrast to models of feedback that incorporate higher order reasoning about mental states (e.g. [9]), this model shows how feedback serves to continually realign processing contexts without recourse to higher order pragmatic reasoning, and provides a mechanistic model of the characteristic divergence and convergence that is key to moving dialogue forward. As well as providing insights into human-human communication, this work has implications for the production and interpretation of human-like feedback in dialogue systems; not just based on unanalysed features (which may result in accurate placement), but because they have successfully compiled a semantic unit. Our FRS model is implemented [10] and deployed in a new dialogue system architecture [11]; ongoing work explores the naturalness and usability of such systems versus, turn-based systems or those without the FRS model. [1] Bavelas, Coates, Johnson (2000). Listeners as co-narrators. Journal of personality and social psychology 79(6). [2] Clark (1996). Using Language. Cambridge University Press. [3] Eshghi, Howes, Hough, Gregoromichelaki, Purver (2015). Feedback in conversation as incremental semantic update. Proceedings IWCS. [4] Howes, Eshghi (2017). Feedback relevance spaces: The organisation of increments in conversation. Proceedings IWCS. [5] Howes, Purver, Healey, Mills, Gregoromichelaki (2011). On incrementality in dialogue: Evidence from compound contributions. Dialogue and Discourse 2(1). [6] Poppe, Truong, Heylen (2011). Backchannels: Quantity, type and timing matters. International Workshop on Intelligent Virtual Agents. [7] Kempson, Cann, Gregoromichelaki, Chatzikiriakidis (2016). Language as mechanisms for interaction. Theoretical Linguistics 42(3-4). [8] Sacks, Schegloff, Jefferson (1974). A simplest systematics for the organization of turntaking for conversation. Language 50(4). [9] Visser, Traum, DeVault, op den Akker (2014). A model for incremental grounding in spoken dialogue systems. Journal on Multimodal User Interfaces 8(1). [10] Eshghi (2015). Dylan: An incremental, semantic, contextual parser for dialogue. Proceedings SEMDIAL. [11] Eshghi, Shalyminov, Lemon. (2017) Bootstrapping incremental dialogue systems from minimal data: linguistic knowledge or machine learning? Proceedings EMNLP.

@inproceedings{
 title = {Bootstrapping dialogue systems: the contribution of a semantic model of interactional dynamics},
 type = {inproceedings},
 year = {2017},
 websites = {https://gupea.ub.gu.se/bitstream/2077/54911/2/gupea_2077_54911_2.pdf},
 id = {1e437bff-b8f3-316e-9657-c9da99f1c66f},
 created = {2019-04-10T10:35:05.695Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-04-10T10:35:05.695Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Eshghi.etal17laml},
 source_type = {inproceedings},
 private_publication = {false},
 bibtype = {inproceedings},
 author = {Eshghi, Arash and Shalyminov, Igor and Lemon, Oliver},
 booktitle = {Proceedings of the Conference on Logic and Machine Learning in Natural Language (LaML)}
}

@inproceedings{
 title = {Bootstrapping incremental dialogue systems from minimal data: the generalisation power of dialogue grammars},
 type = {inproceedings},
 year = {2017},
 websites = {https://arxiv.org/pdf/1709.07858.pdf},
 id = {fe2c2db5-db37-39ea-94d6-39859a926965},
 created = {2019-04-10T10:35:05.708Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-04-10T23:34:37.397Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Eshghi.etal17},
 source_type = {inproceedings},
 private_publication = {false},
 bibtype = {inproceedings},
 author = {Eshghi, Arash and Shalyminov, Igor and Lemon, Oliver},
 booktitle = {Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing (EMNLP)}
}

@inproceedings{
 title = {Feedback Relevance Spaces: the Organisation of Increments in Conversation},
 type = {inproceedings},
 year = {2017},
 websites = {http://www.aclweb.org/anthology/W17-6913},
 id = {a3cbc3cf-f0d3-3905-a9b1-23c9b81a9c0f},
 created = {2019-04-10T10:35:05.767Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-04-10T10:35:05.767Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Howes.Eshghi17},
 source_type = {inproceedings},
 private_publication = {false},
 bibtype = {inproceedings},
 author = {Howes, Christine and Eshghi, Arash},
 booktitle = {Proceedings of the 12th International Conference on Computational Semantics (IWCS)}
}

@inproceedings{
 title = {Interactional Dynamics and the Emergence of Language Games},
 type = {inproceedings},
 year = {2017},
 websites = {http://www.christinehowes.com/papers/fadli/paper_16.pdf},
 city = {Barcelona},
 id = {2b23a4c5-4939-31aa-9bcd-727fdfd07702},
 created = {2019-04-10T10:35:05.890Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-04-10T10:35:05.890Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Eshghi.etal17a},
 source_type = {inproceedings},
 private_publication = {false},
 bibtype = {inproceedings},
 author = {Eshghi, Arash and Shalyminov, Igor and Lemon, Oliver},
 booktitle = {Proceedings of the ESSLLI 2017 workshop on Formal approaches to the Dynamics of Linguistic Interaction}
}

@inproceedings{
 title = {The BURCHAK corpus: a Challenge Data Set for Interactive Learning of Visually Grounded Word Meanings},
 type = {inproceedings},
 year = {2017},
 websites = {https://arxiv.org/pdf/1709.10431.pdf},
 city = {Valencia},
 id = {02538311-89f5-3b5d-99fb-54c14400e46e},
 created = {2019-04-10T10:35:05.957Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-04-10T10:35:05.957Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Yu.etal17},
 source_type = {inproceedings},
 private_publication = {false},
 bibtype = {inproceedings},
 author = {Yu, Yanchao and Eshghi, Arash and Mills, Gregory and Lemon, Oliver},
 booktitle = {Proceedings of the EACL 2017 workshop on Vision and Language (VL'17)}
}

@inproceedings{
 title = {Learning how to learn: an adaptive dialogue agent for incrementally learning visually grounded word meanings},
 type = {inproceedings},
 year = {2017},
 pages = {10-19},
 websites = {http://www.aclweb.org/anthology/W17-2802},
 publisher = {Association for Computational Linguistics},
 id = {903e5ff9-dc72-3c70-8a50-6c6be5c5101f},
 created = {2019-04-10T10:35:06.333Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-04-10T10:35:06.333Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Yu.etal17b},
 source_type = {inproceedings},
 private_publication = {false},
 bibtype = {inproceedings},
 author = {Yu, Yanchao and Eshghi, Arash and Lemon, Oliver},
 booktitle = {Proceedings of the First Workshop on Language Grounding for Robotics}
}

@inproceedings{
 title = {Interactional dynamics and the emergence of language games},
 type = {inproceedings},
 year = {2017},
 pages = {17-21},
 volume = {1863},
 id = {eda151fd-4787-34b7-91a5-84484aebbe0b},
 created = {2019-08-20T09:40:40.087Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2022-06-08T17:58:10.981Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 private_publication = {false},
 abstract = {Meaning is highly activity-specific, in that the action that a particular sequence of words is taken to perform is severely underdetermined in the absence of an overarching activity, or a 'language-game'. In this paper, we combine a formal, incremental model of interactional dynamics and contextual update - Dynamic Syntax and Type Theory with Records (DS-TTR) - with Reinforcement Learning for word selection. We show, using an implemented system, that trial and error generation with a DS-TTR lexicon - a process we have dubbed babbling - leads to particular domain-specific dialogue acts to be learned and routinised over time; and thus that higher level dialogue structures - or how actions fit together to form a coherent whole - can be learned in this fashion. This method therefore allows incremental dialogue systems to be automatically bootstrapped from small amounts of unannotated dialogue transcripts, yet capturing a combinatorially large number of interactional variations. Even when the system is trained from only a single dialogue, we show that it supports over 8000 new dialogues in the same domain. This generalisation property results from the structural knowledge and constraints present within the grammar, and highlights limitations of recent state-of-the-art systems that are built using machine learning techniques only.},
 bibtype = {inproceedings},
 author = {Eshghi, Arash and Shalyminov, Igor and Lemon, Oliver},
 booktitle = {CEUR Workshop Proceedings}
}

Meaning is highly activity-specific, in that the action that a particular sequence of words is taken to perform is severely underdetermined in the absence of an overarching activity, or a 'language-game'. In this paper, we combine a formal, incremental model of interactional dynamics and contextual update - Dynamic Syntax and Type Theory with Records (DS-TTR) - with Reinforcement Learning for word selection. We show, using an implemented system, that trial and error generation with a DS-TTR lexicon - a process we have dubbed babbling - leads to particular domain-specific dialogue acts to be learned and routinised over time; and thus that higher level dialogue structures - or how actions fit together to form a coherent whole - can be learned in this fashion. This method therefore allows incremental dialogue systems to be automatically bootstrapped from small amounts of unannotated dialogue transcripts, yet capturing a combinatorially large number of interactional variations. Even when the system is trained from only a single dialogue, we show that it supports over 8000 new dialogues in the same domain. This generalisation property results from the structural knowledge and constraints present within the grammar, and highlights limitations of recent state-of-the-art systems that are built using machine learning techniques only.

@inproceedings{
 title = {Bootstrapping incremental dialogue systems: using linguistic knowledge to learn from minimal data},
 type = {inproceedings},
 year = {2016},
 websites = {https://arxiv.org/pdf/1612.00347.pdf},
 city = {Barcelona},
 id = {0242da8b-66bc-3181-8531-3e6857a700c1},
 created = {2019-04-10T10:35:05.653Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-04-10T10:35:05.653Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Kalatzis.etal16},
 source_type = {inproceedings},
 private_publication = {false},
 bibtype = {inproceedings},
 author = {Kalatzis, Dimitrios and Eshghi, Arash and Lemon, Oliver},
 booktitle = {Proceedings of the NIPS 2016 workshop on Learning Methods for Dialogue}
}

@inproceedings{
 title = {Comparing dialogue strategies for learning grounded language from human tutors},
 type = {inproceedings},
 year = {2016},
 websites = {http://semantics.rutgers.edu/jersem/proceedings/Semdial_2016_JerSem_proc_full_05_paper_8.pdf},
 city = {New Jersey},
 id = {0b0717a8-aa02-368f-9b6f-6ad907c9b58d},
 created = {2019-04-10T10:35:05.689Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-04-10T10:35:05.689Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Yu.etal16semdial},
 source_type = {inproceedings},
 private_publication = {false},
 bibtype = {inproceedings},
 author = {Yu, Yanchao and Eshghi, Arash and Lemon, Oliver},
 booktitle = {Proceedings of Semdial 2016 (JerSem)}
}

@inproceedings{
 title = {Training an adaptive dialogue policy for interactive learning of visually grounded word meanings},
 type = {inproceedings},
 year = {2016},
 pages = {339-349},
 websites = {http://www.aclweb.org/anthology/W16-3643},
 city = {Los Angeles},
 id = {b37dbb59-d334-3d78-81d7-1fcab145f2ff},
 created = {2019-04-10T10:35:05.970Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-04-10T10:35:05.970Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Yu.etal16},
 source_type = {inproceedings},
 private_publication = {false},
 bibtype = {inproceedings},
 author = {Yu, Yanchao and Eshghi, Arash and Lemon, Oliver},
 booktitle = {Proceedings of SIGDIAL 2016, 17th Annual Meeting of the Special Interest Group on Discourse and Dialogue}
}

@inproceedings{
 title = {An Incremental Dialogue System for Learning Visually Grounded Language (demonstration system)},
 type = {inproceedings},
 year = {2016},
 websites = {http://semantics.rutgers.edu/jersem/proceedings/Semdial_2016_JerSem_proc_short_07_paper_21.pdf},
 city = {New Jersey},
 id = {015e2413-a224-3f39-a256-8dee196618b6},
 created = {2019-04-10T10:35:06.025Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-04-10T10:35:06.025Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Yu.etal16semdemo},
 source_type = {inproceedings},
 private_publication = {false},
 bibtype = {inproceedings},
 author = {Yu, Yanchao and Eshghi, Arash and Lemon, Oliver},
 booktitle = {Proceedings of Semdial 2016 (JerSem)}
}

@inproceedings{
 title = {Incremental Generation of Visually Grounded Language in Situated Dialogue},
 type = {inproceedings},
 year = {2016},
 websites = {http://www.aclweb.org/anthology/W16-6619},
 city = {Los Angeles},
 id = {9176de13-32ea-39ff-bcb0-37eed0a4abeb},
 created = {2019-04-10T10:35:06.195Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-04-10T10:35:06.195Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Yu.etal16inlg},
 source_type = {inproceedings},
 private_publication = {false},
 bibtype = {inproceedings},
 author = {Yu, Yanchao and Eshghi, Arash and Lemon, Oliver},
 booktitle = {Proceedings of INLG 2016}
}

@inproceedings{
 title = {Feedback in Conversation as Incremental Semantic Update},
 type = {inproceedings},
 year = {2015},
 websites = {https://sites.google.com/site/araesh81/eshghi-et-al15-iwcs.pdf},
 id = {789ea54d-e843-301a-91f9-6070ee2e2905},
 created = {2019-04-10T10:35:05.423Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-04-10T10:35:05.423Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {eshghi.etal15},
 source_type = {inproceedings},
 private_publication = {false},
 bibtype = {inproceedings},
 author = {Eshghi, Arash and Howes, Christine and Hough, Julian and Gregoromichelaki, Eleni and Purver, Matthew},
 booktitle = {Proceedings of the 11th International Conference on Computational Semantics (IWCS)}
}

@inproceedings{
 title = {Comparing attribute classifiers for interactive language grounding},
 type = {inproceedings},
 year = {2015},
 websites = {http://www.aclweb.org/anthology/W15-2811},
 id = {c5d74618-3c34-361b-9e91-3a5101b00e83},
 created = {2019-04-10T10:35:05.539Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-04-10T10:35:05.539Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Yu.etal15},
 source_type = {inproceedings},
 private_publication = {false},
 bibtype = {inproceedings},
 author = {Yu, Yanchao and Eshghi, Arash and Lemon, Oliver},
 booktitle = {Proceedings of EMNLP 2015 workshop on Vision and Language}
}

@inproceedings{
 title = {DyLan: An incremental, semantic, contextual parser for dialogue},
 type = {inproceedings},
 year = {2015},
 websites = {https://sites.google.com/site/araesh81/eshghi15-dylan-demo.pdf?attredirects=0&d=1},
 id = {000f1c53-0476-3268-b7d7-98ddc0676b16},
 created = {2019-04-10T10:35:05.574Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-04-10T10:35:05.574Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Eshghi15},
 source_type = {inproceedings},
 private_publication = {false},
 bibtype = {inproceedings},
 author = {Eshghi, Arash},
 booktitle = {Proceedings of SEMDIAL 2015 (GoDIAL)}
}

@article{
 title = {Collective Contexts in Conversation: Grounding by proxy},
 type = {article},
 year = {2015},
 pages = {299-324},
 websites = {https://sites.google.com/site/araesh81/publications/eshghi-healey15a.pdf},
 id = {fba72bf7-07e7-3d68-b40e-a5a42eca01dd},
 created = {2019-04-10T10:35:05.664Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-04-10T10:35:05.664Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Eshghi.Healey15},
 source_type = {article},
 private_publication = {false},
 bibtype = {article},
 author = {Eshghi, Arash and Healey, Patrick G T},
 journal = {Cognitive Science},
 number = {2}
}

@inproceedings{
 title = {Interactive Learning through Dialogue for Multimodal Language Grounding},
 type = {inproceedings},
 year = {2015},
 websites = {http://www.aclweb.org/anthology/W16-3206},
 id = {53e49f65-7667-30f4-bead-9d4ec9c265f2},
 created = {2019-04-10T10:35:05.929Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-04-10T10:35:05.929Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Yu.etal15a},
 source_type = {inproceedings},
 private_publication = {false},
 bibtype = {inproceedings},
 author = {Yu, Yanchao and Lemon, Oliver and Eshghi, Arash},
 booktitle = {Proceedings of SEMDIAL 2015 (GoDIAL)}
}

@inbook{
 type = {inbook},
 year = {2015},
 websites = {https://sites.google.com/site/araesh81/kempson-et-al15hcst.pdf},
 publisher = {Wiley},
 edition = {2nd},
 id = {32af56b1-c3b0-326d-8a60-a21437cdfe85},
 created = {2019-04-10T10:35:06.148Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-04-10T10:35:06.148Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Kempson.etal15},
 source_type = {incollection},
 private_publication = {false},
 bibtype = {inbook},
 author = {Kempson, Ruth and Cann, Ronnie and Eshghi, Arash and Gregoromichelaki, Eleni and Purver, Matthew},
 editor = {Lappin, Shalom and Fox, Chris},
 chapter = {Ellipsis},
 title = {Handbook of Contemporary Semantic Theory}
}

@inproceedings{
 title = {Deep Reinforcement Learning for constructing meaning by `babbling'},
 type = {inproceedings},
 year = {2015},
 websites = {http://www.eecs.qmul.ac.uk/~suew/Publicationsnotindropbox/Purveriwcs2015.pdf},
 id = {439349bf-074c-3408-a8ad-e6087d4ccbdc},
 created = {2019-04-10T10:35:06.272Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-04-10T10:35:06.272Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Lemon.Eshghi15},
 source_type = {inproceedings},
 private_publication = {false},
 bibtype = {inproceedings},
 author = {Lemon, Oliver and Eshghi, Arash},
 editor = {Kempson, Ruth and Cooper, Robin and Purver, Matthew},
 booktitle = {Proceedings of the IMC workshop, IWCS 2016}
}

@inproceedings{
 title = {Feedback in conversation as incremental semantic update},
 type = {inproceedings},
 year = {2015},
 id = {b5bf70e7-64de-31c5-b67a-7aa0e7c361a8},
 created = {2019-08-20T09:40:40.192Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-08-20T09:40:40.192Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {© 2015 Association for Computational Linguistics. In conversation, interlocutors routinely indicate whether something said or done has been processed and integrated. Such feedback includes backchannels such as 'okay' or 'mhm', the production of a next relevant turn, and repair initiation via clarification requests. Importantly, such feedback can be produced not only at sentence/turn boundaries, but also sub-sententially. In this paper, we extend an existing model of incremental semantic processing in dialogue, based around the Dynamic Syntax (DS) grammar framework, to provide a low-level, integrated account of backchannels, clarification requests and their responses; demonstrating that they can be accounted for as part of the core semantic structure-building mechanisms of the grammar, rather than via higher level pragmatic phenomena such as intention recognition, or treatment as an "unofficial" part of the conversation. The end result is an incremental model in which words, not turns, are seen as procedures for contextual update and backchannels serve to align participant semantic processing contexts and thus ease the production and interpretation of subsequent conversational actions. We also show how clarification requests and their following responses and repair can be modelled within the same DS framework, wherein the divergence and re-alignment effort in participants' semantic processing drives conversations forward.},
 bibtype = {inproceedings},
 author = {Eshghi, A. and Howes, C. and Gregoromichelaki, E. and Hough, J. and Purver, M.},
 booktitle = {IWCS 2015 - Proceedings of the 11th International Conference on Computational Semantics}
}

© 2015 Association for Computational Linguistics. In conversation, interlocutors routinely indicate whether something said or done has been processed and integrated. Such feedback includes backchannels such as 'okay' or 'mhm', the production of a next relevant turn, and repair initiation via clarification requests. Importantly, such feedback can be produced not only at sentence/turn boundaries, but also sub-sententially. In this paper, we extend an existing model of incremental semantic processing in dialogue, based around the Dynamic Syntax (DS) grammar framework, to provide a low-level, integrated account of backchannels, clarification requests and their responses; demonstrating that they can be accounted for as part of the core semantic structure-building mechanisms of the grammar, rather than via higher level pragmatic phenomena such as intention recognition, or treatment as an "unofficial" part of the conversation. The end result is an incremental model in which words, not turns, are seen as procedures for contextual update and backchannels serve to align participant semantic processing contexts and thus ease the production and interpretation of subsequent conversational actions. We also show how clarification requests and their following responses and repair can be modelled within the same DS framework, wherein the divergence and re-alignment effort in participants' semantic processing drives conversations forward.

@inproceedings{
 title = {How domain-general can we be? Learning incremental dialogue systems without dialogue acts},
 type = {inproceedings},
 year = {2014},
 websites = {https://sites.google.com/site/araesh81/publications/eshghi-lemon-14.pdf},
 id = {8c55210a-1927-3519-8635-305ebc33ea36},
 created = {2019-04-10T10:35:05.686Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-04-10T10:35:05.686Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Eshghi.Lemon14},
 source_type = {inproceedings},
 private_publication = {false},
 bibtype = {inproceedings},
 author = {Eshghi, Arash and Lemon, Oliver},
 booktitle = {Proceedings of SEMDIAL}
}

@inproceedings{
 title = {Making things worse to make them better: The role of negative evidence in the coordination of referring expressions},
 type = {inproceedings},
 year = {2013},
 websites = {http://homepages.inf.ed.ac.uk/gmills/HealeyMillsEshghi2013.pdf},
 id = {ba5335f6-c5bf-32b2-a0ba-4f24daa0ae61},
 created = {2019-04-10T10:35:05.775Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-04-10T10:35:05.775Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Healey.etal13},
 source_type = {inproceedings},
 private_publication = {false},
 bibtype = {inproceedings},
 author = {Healey, Patrick G T and Mills, Gregory J and Eshghi, Arash},
 booktitle = {Proceedings of 35th Annual Meeting of the Cognitive Science Society (CogSci 2013)}
}

@inproceedings{
 title = {Incremental Grammar Induction from Child-Directed Dialogue Utterances},
 type = {inproceedings},
 year = {2013},
 pages = {94-103},
 websites = {http://www.aclweb.org/anthology/W13-2611},
 month = {8},
 publisher = {Association for Computational Linguistics},
 city = {Sofia, Bulgaria},
 id = {9256185e-4d79-3b35-87d1-6f873f392777},
 created = {2019-04-10T10:35:05.992Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-04-10T10:35:05.992Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Eshghi.etal13a},
 source_type = {inproceedings},
 private_publication = {false},
 bibtype = {inproceedings},
 author = {Eshghi, Arash and Hough, Julian and Purver, Matthew},
 booktitle = {Proceedings of the 4th Annual Workshop on Cognitive Modeling and Computational Linguistics (CMCL)}
}

@article{
 title = {``Well that's \emphone way'': Interactivity in parsing and production (commentary on Pickering and Garrod's ``An integrated theory of language production and comprehension'')},
 type = {article},
 year = {2013},
 websites = {https://www.cambridge.org/core/journals/behavioral-and-brain-sciences/article/well-thats-one-way-interactivity-in-parsing-and-production/DF33EA13EAFB8DCA50A5791127E7700F},
 id = {1ef33328-567d-37d4-8fc2-6934b3fd4cb9},
 created = {2019-04-10T10:35:06.042Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-04-10T10:35:06.042Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Howes.etal13b},
 source_type = {article},
 private_publication = {false},
 bibtype = {article},
 author = {Howes, Christine and Healey, Patrick G T and Eshghi, Arash and Hough, Julian},
 journal = {Behavioral and Brain Sciences}
}

@inproceedings{
 title = {Probabilistic induction for an incremental semantic grammar},
 type = {inproceedings},
 year = {2013},
 pages = {107-118},
 websites = {http://www.aclweb.org/anthology/W13-0110},
 month = {3},
 publisher = {Association for Computational Linguistics},
 city = {Potsdam, Germany},
 id = {e45691ba-88b1-346e-becd-4f24b7e11c03},
 created = {2019-04-10T10:35:06.042Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-04-10T10:35:06.042Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Eshghi.etal13},
 source_type = {inproceedings},
 private_publication = {false},
 bibtype = {inproceedings},
 author = {Eshghi, Arash and Purver, Matthew and Hough, Julian},
 booktitle = {Proceedings of the 10th International Conference on Computational Semantics (IWCS 2013) -- Long Papers}
}

@inbook{
 type = {inbook},
 year = {2013},
 websites = {http://www.dynamicsyntax.org/papers/GregoromichelakiEtAlOnMakingSyntaxDynamic2013Draft.pdf},
 publisher = {John Benjamins Publishing Company},
 id = {850a2fe2-3f45-374a-8781-acae28f850e2},
 created = {2019-04-10T10:35:06.070Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-04-10T10:35:06.070Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Gregoromichelaki.etal13},
 source_type = {incollection},
 notes = {Series: Advances in Interaction Studies (series editors: Kerstin Dautenhahn, Angelo Cangelosi)},
 private_publication = {false},
 bibtype = {inbook},
 author = {Gregoromichelaki, Eleni and Kempson, Ruth and Howes, Christine and Eshghi, Arash},
 editor = {Wachsmuth, Ipke and de Ruiter, Jan and Jaecks, Petra and Kopp, Stefan},
 chapter = {On making syntax dynamic: the challenge of compound utterances and the architecture of the grammar},
 title = {Alignment in Communication: Towards a New Theory of Communication}
}

@article{
 title = {Well, that's one way: Interactivity in parsing and production},
 type = {article},
 year = {2013},
 volume = {36},
 id = {83f12360-9b9e-3f39-aa32-362fbc262e37},
 created = {2019-08-20T09:40:40.148Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-08-20T09:40:40.148Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {We present empirical evidence from dialogue that challenges some of the key assumptions in the Pickering & Garrod (P&G) model of speaker-hearer coordination in dialogue. The P&G model also invokes an unnecessarily complex set of mechanisms. We show that a computational implementation, currently in development and based on a simpler model, can account for more of this type of dialogue data. Copyright © 2013 Cambridge University Press.},
 bibtype = {article},
 author = {Howes, C. and Healey, P.G.T. and Eshghi, A. and Hough, J.},
 doi = {10.1017/S0140525X12002592},
 journal = {Behavioral and Brain Sciences},
 number = {4}
}

We present empirical evidence from dialogue that challenges some of the key assumptions in the Pickering & Garrod (P&G) model of speaker-hearer coordination in dialogue. The P&G model also invokes an unnecessarily complex set of mechanisms. We show that a computational implementation, currently in development and based on a simpler model, can account for more of this type of dialogue data. Copyright © 2013 Cambridge University Press.

@inbook{
 type = {inbook},
 year = {2012},
 websites = {https://sites.google.com/site/araesh81/eshghi-et-al-12-ds-ttr.pdf},
 publisher = {College Publications},
 id = {991b6e8f-868e-3890-a67b-7015ff2a9d17},
 created = {2019-04-10T10:35:05.795Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-04-10T10:35:05.795Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Eshghi.etal12b},
 source_type = {incollection},
 private_publication = {false},
 bibtype = {inbook},
 author = {Eshghi, Arash and Hough, Julian and Purver, Matthew and Kempson, Ruth and Gregoromichelaki, Eleni},
 chapter = {Conversational Interactions: Capturing Dialogue Dynamics},
 title = {From Quantification to Conversation: Festschrift for Robin Cooper on the occasion of his 65th birthday}
}

@inproceedings{
 title = {Finishing each other's ... Responding to incomplete contributions in dialogue},
 type = {inproceedings},
 year = {2012},
 pages = {479-484},
 websites = {https://mindmodeling.org/cogsci2012/papers/0094/paper0094.pdf},
 month = {8},
 city = {Sapporo, Japan},
 id = {26b2d196-384c-3c81-9caf-af51776a1f4f},
 created = {2019-04-10T10:35:05.881Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-04-10T10:35:05.881Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Howes.etal12b},
 source_type = {inproceedings},
 notes = {ISBN 978-0-9768318-8-4},
 private_publication = {false},
 bibtype = {inproceedings},
 author = {Howes, Christine and Healey, Patrick G T and Purver, Matthew and Eshghi, Arash},
 booktitle = {Proceedings of the 34th Annual Meeting of the Cognitive Science Society (CogSci 2012)}
}

@inproceedings{
 title = {Inducing Lexical Entries in an Incremental Semantic Grammar},
 type = {inproceedings},
 year = {2012},
 websites = {https://sites.google.com/site/araesh81/EshghiEtAl12CSLP.pdf},
 id = {4ed9ace6-02e1-310b-be77-5e5f13b90cd9},
 created = {2019-04-10T10:35:05.906Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-04-10T10:35:05.906Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Eshghi.etal12a},
 source_type = {inproceedings},
 private_publication = {false},
 bibtype = {inproceedings},
 author = {Eshghi, Arash and Purver, Matthew and Hough, Julian and Sato, Yo},
 booktitle = {Proceedings of CSLP2012}
}

@inproceedings{
 title = {Non-verbal cues to recipient roles in dialogue},
 type = {inproceedings},
 year = {2011},
 id = {2f76d307-6ee0-35c7-897d-1e482dfc4702},
 created = {2019-04-10T10:35:05.814Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-04-10T10:35:05.814Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Battersby.etal11},
 source_type = {inproceedings},
 private_publication = {false},
 bibtype = {inproceedings},
 author = {Battersby, Stuart A and Healey, Patrick G T and Lavelle, Mary and Eshghi, Arash and McCabe, Rose},
 booktitle = {Proceedings of the 21st annual meeting of the Society for Text and Discourse}
}

@techreport{
 title = {Dylan: Parser For Dynamic Syntax},
 type = {techreport},
 year = {2011},
 websites = {https://drive.google.com/open?id=0B7DgzzsvsmBwMXdfbmp5ak0xMUE},
 institution = {Queen Mary University of London},
 id = {0d118875-924c-3a7c-8eec-051a986bac35},
 created = {2019-04-10T10:35:05.814Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-06-16T14:55:13.849Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Eshghi.etal11},
 source_type = {techreport},
 private_publication = {false},
 bibtype = {techreport},
 author = {Eshghi, A and Purver, M and Hough, Julian}
}

@inproceedings{
 title = {Making a Contribution: Processing clarification requests in dialogue},
 type = {inproceedings},
 year = {2011},
 websites = {http://www.eecs.qmul.ac.uk/~mpurver/papers/healey-et-al11std.pdf},
 month = {7},
 city = {Poitiers},
 id = {1ed0de57-d11b-3997-bdb1-acf108d30a1f},
 created = {2019-04-10T10:35:06.252Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-04-10T10:35:06.252Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Healey.etal11},
 source_type = {inproceedings},
 private_publication = {false},
 bibtype = {inproceedings},
 author = {Healey, P G T and Eshghi, Arash and Howes, Christine and Purver, Matthew},
 booktitle = {Proceedings of the 21st Annual Meeting of the Society for Text and Discourse}
}

@inproceedings{
 title = {Incremental Semantic Construction in a Dialogue System},
 type = {inproceedings},
 year = {2011},
 pages = {365-369},
 websites = {https://pdfs.semanticscholar.org/a5b7/ff9bb733a0eb9fd9ec67886fd6391565d827.pdf},
 month = {1},
 city = {Oxford, UK},
 id = {c6b536d9-bff4-344b-88ac-0ed66c462ba2},
 created = {2019-04-10T10:35:06.255Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-04-10T10:35:06.255Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Purver.etal11},
 source_type = {inproceedings},
 private_publication = {false},
 bibtype = {inproceedings},
 author = {Purver, Matthew and Eshghi, Arash and Hough, Julian},
 editor = {Bos, J and Pulman, S},
 booktitle = {Proceedings of the 9th International Conference on Computational Semantics}
}

@inproceedings{
 title = {Triangulations: Simultaneous engagement in multi-party interaction},
 type = {inproceedings},
 year = {2010},
 pages = {255},
 id = {43477619-9be6-3a44-a761-94763139e015},
 created = {2019-04-10T10:35:05.814Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-04-10T10:35:05.814Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Battersby.etal10},
 source_type = {inproceedings},
 private_publication = {false},
 bibtype = {inproceedings},
 author = {Battersby, Stuart A and Healey, Patrick G T and Eshghi, Arash},
 editor = {Deppermann, A and Spranz-Fogasy, T},
 booktitle = {Proceedings of the International Conference on Conversation Analysis}
}

@inproceedings{
 title = {Incremental Turn Processing in Dialogue},
 type = {inproceedings},
 year = {2010},
 websites = {http://www.eecs.qmul.ac.uk/~mpurver/papers/eshghi-et-al10amlap.pdf},
 month = {9},
 city = {York, UK},
 id = {cce9d798-7af1-3f3d-ba4e-a57e57bb2e94},
 created = {2019-04-10T10:35:06.130Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-04-10T10:35:06.130Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Eshghi.etal10},
 source_type = {inproceedings},
 private_publication = {false},
 bibtype = {inproceedings},
 author = {Eshghi, Arash and Healey, P G T and Purver, Matthew and Howes, Christine and Gregoromichelaki, Eleni and Kempson, Ruth},
 booktitle = {Architectures and Mechanisms for Language Processing}
}

@inproceedings{
 title = {What is Conversation? Distinguishing Dialogue Contexts},
 type = {inproceedings},
 year = {2009},
 websites = {https://sites.google.com/site/araesh81/publications/eshghi-healey09.pdf},
 id = {3e68e56a-5504-36ba-8302-162fad9a5d70},
 created = {2019-04-10T10:35:06.154Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-04-10T10:35:06.154Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Eshghi.Healey09},
 source_type = {inproceedings},
 private_publication = {false},
 bibtype = {inproceedings},
 author = {Eshghi, Arash and Healey, P G T},
 editor = {Taatgen, Niels and van Rijn, Hedderik},
 booktitle = {Proceedings of the Thirty-First Annual Conference of the Cognitive Science Society}
}

@phdthesis{
 title = {Uncommon Ground: The Distribution of Dialogue Contexts},
 type = {phdthesis},
 year = {2009},
 websites = {https://sites.google.com/site/araesh81/publications/thesis.pdf},
 institution = {Queen Mary University of London},
 id = {ce7a113c-26fa-3e1c-b78c-e72c740045e9},
 created = {2019-04-10T10:35:06.185Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-04-10T10:35:06.185Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Eshghi09},
 source_type = {phdthesis},
 private_publication = {false},
 bibtype = {phdthesis},
 author = {Eshghi, Arash}
}

@article{
 title = {Communication Spaces},
 type = {article},
 year = {2008},
 websites = {https://sites.google.com/site/araesh81/healey-et-al-08-com-spaces.pdf},
 id = {6c7b956c-140f-3ae4-8380-777033531352},
 created = {2019-04-10T10:35:05.423Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-04-10T10:35:05.423Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Healey.etal08},
 source_type = {article},
 user_context = {academic},
 notes = {Online first: http://www.springerlink.com/content/0925-9724},
 private_publication = {false},
 bibtype = {article},
 author = {Healey, P G T and White, G and Eshghi, A and Reeves, A and Light, A},
 doi = {10.1007/s10606-007-9061-4},
 journal = {Computer Supported Co-operative Work}
}

@inproceedings{
 title = {Quantifying Ellipsis in Dialogue: An index of mutual understanding},
 type = {inproceedings},
 year = {2008},
 websites = {http://www.aclweb.org/anthology/W08-0116},
 publisher = {Association for Computational Linguistics},
 id = {ca00aabe-4fde-30ad-9f8f-14e4fc78050f},
 created = {2019-04-10T10:35:06.292Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-04-10T10:35:06.292Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Colman.etal08},
 source_type = {inproceedings},
 private_publication = {false},
 bibtype = {inproceedings},
 author = {Colman, M and Eshghi, A and Healey, P G T},
 booktitle = {Proceedings of SIGDIAL 2008}
}

@inproceedings{
 title = {Group Dialects in an Online Community},
 type = {inproceedings},
 year = {2007},
 websites = {http://events.illc.uva.nl/semdial/proceedings/semdial2007_decalog_proceedings.pdf},
 id = {bf0fcd7f-dea3-3eba-952c-4feb5e0c956b},
 created = {2019-04-10T10:35:05.542Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-04-10T10:35:05.542Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Healey.etal07b},
 source_type = {inproceedings},
 private_publication = {false},
 bibtype = {inproceedings},
 author = {Healey, Patrick G T and Vogel, Carl and Eshghi, Arash},
 booktitle = {Proceedings of Decalog 2007, the 11th Workshop on the Semantics and Pragmatics of Dialogue}
}

@inproceedings{
 title = {Collective States of Understanding},
 type = {inproceedings},
 year = {2007},
 pages = {2-9},
 websites = {https://sites.google.com/site/araesh81/eshghi-healey07-sigdial.pdf},
 publisher = {Association for Computational Linguisitics},
 id = {a943586b-5faa-3ddc-9991-57aebc40f32d},
 created = {2019-04-10T10:35:06.082Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-04-10T10:35:06.082Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Eshghi.Healey07},
 source_type = {inproceedings},
 private_publication = {false},
 bibtype = {inproceedings},
 author = {Eshghi, A and Healey, P G T},
 editor = {Keizer, S and Bunt, H and Paek, T},
 booktitle = {Proceedings of the 8th SIGdial Workshop on Discourse and Dialogue}
}

@inproceedings{
 title = {Collective states of understanding},
 type = {inproceedings},
 year = {2007},
 id = {e2451614-1957-356c-8b87-19486cb6d3ff},
 created = {2019-08-20T09:40:40.087Z},
 file_attached = {false},
 profile_id = {d7d2e6da-aa5b-3ab3-b3f2-a5350adf574a},
 last_modified = {2019-08-20T09:40:40.087Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {This paper uses an analysis of ellipsis in multi-party interaction to investigate the relative accessibility of dialogue context/common ground to direct addressees and side participants. The results show that side-participants frequently make direct use of the common ground established between a speaker and addressee despite the fact that, by definition, they did not directly collaborate with the speaker on constructing it. Different individuals can thus reach the same level of grounding through different levels of feedback. We conclude that mutliparty dialogue leads to distinct collective states of understanding that are not reducible to the component dyadic interactions. © 2007 Association for Computational Linguistics.},
 bibtype = {inproceedings},
 author = {Eshghi, A. and Healey, P.G.T.},
 booktitle = {Proceedings of the 8th SIGdial Workshop on Discourse and Dialogue}
}

This paper uses an analysis of ellipsis in multi-party interaction to investigate the relative accessibility of dialogue context/common ground to direct addressees and side participants. The results show that side-participants frequently make direct use of the common ground established between a speaker and addressee despite the fact that, by definition, they did not directly collaborate with the speaker on constructing it. Different individuals can thus reach the same level of grounding through different levels of feedback. We conclude that mutliparty dialogue leads to distinct collective states of understanding that are not reducible to the component dyadic interactions. © 2007 Association for Computational Linguistics.