Skip to main content

Matthew Purver

Queen Mary, University of London, School of Electronic Engineering and Computer Science, Faculty Member

Followers

116

Following

30

Co-authors

27

Public Views

Researcher in computational linguistics / natural language processing, particularly for conversational dialogue. Reader in Computational Linguistics in the School of Electronic Engineering and Computer Science at Queen Mary University of London. Joint head of the Computational Linguistics Lab and deputy head of the Cognitive Science research group.
Address: London, United Kingdom

less

Vrije Universiteit Brussel

University of California, Merced

University of Cambridge

Microsoft Research

The University of Sydney

Kansas State University

Louis de Saussure

University of Neuchâtel

Armando Marques-Guedes

UNL - New University of Lisbon

Ucla

The Hebrew University of Jerusalem

InterestsView All (13)

Uploads

Papers by Matthew Purver

Tracing Linguistic Markers of Influence in a Large Online Organisation

Social science and psycholinguistic research have shown that power and status affect how people u... more Social science and psycholinguistic research have shown that power and status affect how people use language in a range of domains. Here, we investigate a similar question in a large, distributed, consensus-driven community -the Internet Engineering Task Force (IETF), a collaborative organisation that develops technical standards for the Internet. Our analysis, based on lexical categories (LIWC) and BERT, shows that participants' levels of influence can be predicted from their email text, and identifies key linguistic differences (e.g., certain LIWC categories, such as WE are positively correlated with high-influence). We also identify the differences in language use for the same person before and after becoming influential 1 .

LEDA: a Large-Organization Email-Based Decision-Dialogue-Act Analysis Dataset

Collaboration increasingly happens online. This is especially true for large groups working on gl... more Collaboration increasingly happens online. This is especially true for large groups working on global tasks, with collaborators all around the world. The size and distributed nature of such groups make decision-making challenging. This paper proposes a set of dialog acts for the study of decision-making mechanisms in such groups, and provides a new annotated dataset based on real-world data from the public mail-archives of one such organization -the Internet Engineering Task Force (IETF). We provide an initial data analysis showing that this dataset can be used to better understand decision-making in such organizations. Finally, we experiment with a preliminary transformerbased dialog act tagging model.

Proceedings of the Fourth {SIGdial} Workshop on Discourse and Dialogue

Experimenting with Clarification in Dialogue

A new technique for integrating experimental manipulations into text-based, synchronous dialogue ... more A new technique for integrating experimental manipulations into text-based, synchronous dialogue is introduced. This method supports fine-grained, systematic transformation of conversational turns and the introduction of 'artificial' probe turns and turn sequences. It can be used to introduce manipulations that are sensitive to aspects of the local linguistic and conversational context for any task or dialogue type. The use of this technique is illustrated by an experimental investigation of the effect of word category and level of grounding on the interpretation of reprise clarifications. The results show that these factors affect both the type and likelihood of response to reprise fragment clarifications.

Incremental semantic construction in a dialogue system

This paper describes recent work on the DynDial project * towards incremental semantic interpreta... more This paper describes recent work on the DynDial project * towards incremental semantic interpretation in dialogue. We outline our domain-general grammar-based approach, using a variant of Dynamic Syntax integrated with Type Theory with Records and a Davidsonian event-based semantics. We describe a Java-based implementation of the parser, used within the Jindigo framework to produce an incremental dialogue system capable of handling inherently incremental phenomena such as split utterances, adjuncts, and mid-sentence clarification requests or backchannels. * The authors were supported by the Dynamics of Conversational Dialogue project (DynDial -ESRC-RES-062-23-0962). We thank Shalom Lappin, Tim Fernando, Yo Sato, our project colleagues and the anonymous reviewers for helpful comments.

Evaluating Neural Word Representations in Tensor-Based Compositional Settings

arXiv (Cornell University), Aug 26, 2014

We provide a comparative study between neural word representations and traditional vector spaces ... more We provide a comparative study between neural word representations and traditional vector spaces based on cooccurrence counts, in a number of compositional tasks. We use three different semantic spaces and implement seven tensor-based compositional models, which we then test (together with simpler additive and multiplicative approaches) in tasks involving verb disambiguation and sentence similarity. To check their scalability, we additionally evaluate the spaces using simple compositional methods on larger-scale tasks with less constrained language: paraphrase detection and dialogue act tagging. In the more constrained tasks, co-occurrence vectors are competitive, although choice of compositional method is important; on the largerscale tasks, they are outperformed by neural word embeddings, which show robust, stable performance across the tasks.

Interactive Meaning Construction A Workshop at IWCS 2015 Proceedings

Robust Co-occurrence Quantification for Lexical Distributional Semantics

Previous optimisations of parameters affecting the word-context association measure used in distr... more Previous optimisations of parameters affecting the word-context association measure used in distributional vector space models have focused either on highdimensional vectors with hundreds of thousands of dimensions, or dense vectors with dimensionality of a few hundreds; but dimensionality of a few thousands is often applied in compositional tasks as it is still computationally feasible and does not require the dimensionality reduction step. We present a systematic study of the interaction of the parameters of the association measure and vector dimensionality, and derive parameter selection heuristics that achieve performance across word similarity and relevance datasets competitive with the results previously reported in the literature achieved by highly dimensional or dense models.

Alzheimer’s Dementia Recognition From Spontaneous Speech Using Disfluency and Interactional Features

Frontiers in computer science, Jun 18, 2021

Alzheimer's disease (AD) is a progressive, neurodegenerative disorder mainly characterized by mem... more Alzheimer's disease (AD) is a progressive, neurodegenerative disorder mainly characterized by memory loss with deficits in other cognitive domains, including language, visuospatial abilities, and changes in behavior. Detecting diagnostic biomarkers that are noninvasive and cost-effective is of great value not only for clinical assessments and diagnostics but also for research purposes. Several previous studies have investigated AD diagnosis via the acoustic, lexical, syntactic, and semantic aspects of speech and language. Other studies include approaches from conversation analysis that look at more interactional aspects, showing that disfluencies such as fillers and repairs, and purely nonverbal features such as inter-speaker silence, can be key features of AD conversations. These kinds of features, if useful for diagnosis, may have many advantages: They are simple to extract and relatively language-, topic-, and taskindependent. This study aims to quantify the role and contribution of these features of interaction structure in predicting whether a dialogue participant has AD. We used a subset of the Carolinas Conversation Collection dataset of patients with AD at moderate stage within the age range 60-89 and similar-aged non-AD patients with other health conditions. Our feature analysis comprised two sets: disfluency features, including indicators such as self-repairs and fillers, and interactional features, including overlaps, turn-taking behavior, and distributions of different types of silence both within patient speech and between patient and interviewer speech. Statistical analysis showed significant differences between AD and non-AD groups for several disfluency features (edit terms, verbatim repeats, and substitutions) and interactional features (lapses, gaps, attributable silences, turn switches per minute, standardized phonation time, and turn length). For the classification of AD patient conversations vs. non-AD patient conversations, we achieved 83% accuracy with disfluency features, 83% accuracy with interactional features, and an overall accuracy of 90% when combining both feature sets using support vector machine classifiers. The discriminative power of these features, perhaps combined with more conventional linguistic features, therefore shows potential for integration into noninvasive clinical assessments for AD at advanced stages.

On the Means for Clarification in Dialogue

Springer eBooks, 2003

The ability to request clarification of utterances is a vital part of the communicative process. ... more The ability to request clarification of utterances is a vital part of the communicative process. In this paper we discuss the range of possible forms for clarification requests, together with the range of readings they can convey. We present the results of corpus analysis which show a correlation between certain forms and possible readings, together with some indication of maximum likely distance between request and the utterance being clarified. We then explain the implications of these results for a possible HPSG analysis of clarification requests and for an ongoing implementation of a clarification-capable dialogue system.

The Web We Weave: Untangling the Social Graph of the IETF

Proceedings of the International AAAI Conference on Web and Social Media, May 31, 2022

The Internet Engineering Task Force (IETF) has developed many of the technical standards that und... more The Internet Engineering Task Force (IETF) has developed many of the technical standards that underpin the Internet. The standards development process followed by the IETF is open and consensus-driven, but is inherently both a social and political activity, and latent influential structures might exist within the community. Exploring and understanding these structures is essential to ensuring the IETF's resilience and openness. We use network analysis to explore the social graph of IETF participants, based on public email discussions and co-author relationships, and the influence of key contributors. We show that a small core of participants dominates: the top 10% contribute almost half (43.75%) of the emails and come from a relatively small group of organisations. On the other hand, we also find that influence has become relatively more decentralised with time. IETF participants also propose and work on drafts that are either adopted by a working group for further refinement or get rejected at an early stage. Using the social graph features combined with email text features, we perform regression analysis to understand the effect of user influence on the success of new work being adopted by the IETF. Our findings shed useful insights into the behavior of participants across time, correlation between influence and success in draft adoption, and the significance of affiliated organisations in the authorship of drafts.

Ludics, Dialogue and Interaction

The use of general descriptive names, registered names, trademarks, etc. in this publication does... more

Lattice Theoretic Relevance in Incremental Reference Processing

Incrementality, Alignment and Shared Utterances

Institute of Philosophy, 2004

This paper describes an implemented prototype dialogue model within the Dynamic Syntax (DS) frame... more This paper describes an implemented prototype dialogue model within the Dynamic Syntax (DS) framework which directly reflects dialogue phenomena such as alignment, routinization and shared utterances. In DS, word-by-word incremental parsing and generation are defined in terms of actions on semantic tree structures. This paper proposes a model of dialogue context which includes these trees and their associated actions, and shows how alignment and routinization result directly from minimisation of lexicon search (and hence speaker's effort), and how switch of speaker/hearer roles in shared utterances can be seen as a switch between incremental processes directed by different goals, but sharing the same (partial) data structures. 1 Example (1) from the BNC, file KNY (sentences 315-317). 2 Although see for an initial DRT-based approach.

Probabilistic Record Type Lattices for Incremental Reference Processing

Studies in linguistics and philosophy, 2017

Completability vs (In)completeness

Acta Linguistica Hafniensia, 2020

In natural conversation, no notion of "complete sentence" is required for syntactic licensing. Bu... more In natural conversation, no notion of "complete sentence" is required for syntactic licensing. But so-called "fragmentary", "incomplete", and abandoned utterances constitute problematic data for standard formalisms. We argue, instead, that such data contextualised show that: (a) non-sentential utterances are adequate to underpin people's coordination, while (b) all linguistic dependencies can be systematically distributed across more than one participant and turn. Standard models have problems accounting for such data because their notions of 'constituency' and 'syntactic domain' are independent of performance considerations. Concomitantly, no notion of "full proposition" or encoded speech act is necessary for successful interaction. Strings, contents, and joint actions emerge incrementally 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 online, at each step affording and constraining possibilities for further action. For this reason, they need to be modelled as such: a representational level of abstract syntax, divorced from conceptual structure and physical action, impedes a natural account of subsentential coordination phenomena. Instead, we argue that we need a view of grammar as a "skill" employing domain-general mechanisms rather than fixed form-meaning mappings. 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".

Information-Theoretic Segmentation of Natural Language

We present computational experiments on language segmentation using a general information-theoret... more We present computational experiments on language segmentation using a general information-theoretic cognitive model. We present a method which uses the statistical regularities of language to segment a continuous stream of symbols into “meaningful units” at a range of levels. Given a string of symbols—in the present approach, textual representations of phonemes—we attempt to find the syllables such as grea and sy (in the word greasy); words such as in, greasy, wash, and water ; and phrases such as in greasy wash water. The approach is entirely information-theoretic, and requires no knowledge of the units themselves; it is thus assumed to require only general cognitive abilities, and has previously been applied to music. We tested our approach on two spoken language corpora, and we discuss our results in the context of learning as a statistical processes.

Exploring Semantic Incrementality with Dynamic Syntax and Vector Space Semantics

ArXiv, 2018

One of the fundamental requirements for models of semantic processing in dialogue is incrementali... more One of the fundamental requirements for models of semantic processing in dialogue is incrementality: a model must reflect how people interpret and generate language at least on a word-by-word basis, and handle phenomena such as fragments, incomplete and jointly-produced utterances. We show that the incremental word-by-word parsing process of Dynamic Syntax (DS) can be assigned a compositional distributional semantics, with the composition operator of DS corresponding to the general operation of tensor contraction from multilinear algebra. We provide abstract semantic decorations for the nodes of DS trees, in terms of vectors, tensors, and sums thereof; using the latter to model the underspecified elements crucial to assigning partial representations during incremental processing. As a working example, we give an instantiation of this theory using plausibility tensors of compositional distributional semantics, and show how our framework can incrementally assign a semantic plausibilit...

Rare-Class Dialogue Act Tagging for Alzheimer’s Disease Diagnosis

Alzheimer’s Disease (AD) is associated with many characteristic changes, not only in an individua... more Alzheimer’s Disease (AD) is associated with many characteristic changes, not only in an individual’s language but also in the interactive patterns observed in dialogue. The most indicative changes of this latter kind tend to be associated with relatively rare dialogue acts (DAs), such as those involved in clarification exchanges and responses to particular kinds of questions. However, most existing work in DA tagging focuses on improving average performance, effectively prioritizing more frequent classes; it thus gives a poor performance on these rarer classes and is not suited for application to AD analysis. In this paper, we investigate tagging specifically for rare class DAs, using a hierarchical BiLSTM model with various ways of incorporating information from previous utterances and DA tags in context. We show that this can give good performance for rare DA classes on both the general Switchboard corpus (SwDA) and an AD-specific conversational dataset, the Carolinas Conversation...

Shifting Opinions: Experiments on Agreement and Disagreement in Dialogue

Disagreement is understood to be socially problematic; it also rarely surfaces in naturally occur... more Disagreement is understood to be socially problematic; it also rarely surfaces in naturally occurring conversation. An experiment was designed to allow us to directly manipulate the occurrence of exposed (dis)agreement and track its effects on the subsequent dialogue. This is the first experiment to directly manipulate the occurrence of exposed agreement and disagreement in dialogue. Insertions of exposed disagreement disrupt dialogues, bringing the topic of disagreement directly into the conversation, provoking clarification requests and resulting in a greater number of self-edits when formulating turns. The insertion of disagreement also led to more instances of exposed agreement, suggesting that dialogue partners co-operate to redress the face-threat of disagreement. Conversely, exposed agreement insertions were not as incongruous and had less disruptive impact on the ensuing dialogues; however, introducing agreement into the dialogue did lead to greater deliberation, with more a...

Tracing Linguistic Markers of Influence in a Large Online Organisation

Social science and psycholinguistic research have shown that power and status affect how people u... more Social science and psycholinguistic research have shown that power and status affect how people use language in a range of domains. Here, we investigate a similar question in a large, distributed, consensus-driven community -the Internet Engineering Task Force (IETF), a collaborative organisation that develops technical standards for the Internet. Our analysis, based on lexical categories (LIWC) and BERT, shows that participants' levels of influence can be predicted from their email text, and identifies key linguistic differences (e.g., certain LIWC categories, such as WE are positively correlated with high-influence). We also identify the differences in language use for the same person before and after becoming influential 1 .

LEDA: a Large-Organization Email-Based Decision-Dialogue-Act Analysis Dataset

Collaboration increasingly happens online. This is especially true for large groups working on gl... more Collaboration increasingly happens online. This is especially true for large groups working on global tasks, with collaborators all around the world. The size and distributed nature of such groups make decision-making challenging. This paper proposes a set of dialog acts for the study of decision-making mechanisms in such groups, and provides a new annotated dataset based on real-world data from the public mail-archives of one such organization -the Internet Engineering Task Force (IETF). We provide an initial data analysis showing that this dataset can be used to better understand decision-making in such organizations. Finally, we experiment with a preliminary transformerbased dialog act tagging model.

Proceedings of the Fourth {SIGdial} Workshop on Discourse and Dialogue

Experimenting with Clarification in Dialogue

A new technique for integrating experimental manipulations into text-based, synchronous dialogue ... more A new technique for integrating experimental manipulations into text-based, synchronous dialogue is introduced. This method supports fine-grained, systematic transformation of conversational turns and the introduction of 'artificial' probe turns and turn sequences. It can be used to introduce manipulations that are sensitive to aspects of the local linguistic and conversational context for any task or dialogue type. The use of this technique is illustrated by an experimental investigation of the effect of word category and level of grounding on the interpretation of reprise clarifications. The results show that these factors affect both the type and likelihood of response to reprise fragment clarifications.

Incremental semantic construction in a dialogue system

This paper describes recent work on the DynDial project * towards incremental semantic interpreta... more This paper describes recent work on the DynDial project * towards incremental semantic interpretation in dialogue. We outline our domain-general grammar-based approach, using a variant of Dynamic Syntax integrated with Type Theory with Records and a Davidsonian event-based semantics. We describe a Java-based implementation of the parser, used within the Jindigo framework to produce an incremental dialogue system capable of handling inherently incremental phenomena such as split utterances, adjuncts, and mid-sentence clarification requests or backchannels. * The authors were supported by the Dynamics of Conversational Dialogue project (DynDial -ESRC-RES-062-23-0962). We thank Shalom Lappin, Tim Fernando, Yo Sato, our project colleagues and the anonymous reviewers for helpful comments.

Evaluating Neural Word Representations in Tensor-Based Compositional Settings

arXiv (Cornell University), Aug 26, 2014

We provide a comparative study between neural word representations and traditional vector spaces ... more We provide a comparative study between neural word representations and traditional vector spaces based on cooccurrence counts, in a number of compositional tasks. We use three different semantic spaces and implement seven tensor-based compositional models, which we then test (together with simpler additive and multiplicative approaches) in tasks involving verb disambiguation and sentence similarity. To check their scalability, we additionally evaluate the spaces using simple compositional methods on larger-scale tasks with less constrained language: paraphrase detection and dialogue act tagging. In the more constrained tasks, co-occurrence vectors are competitive, although choice of compositional method is important; on the largerscale tasks, they are outperformed by neural word embeddings, which show robust, stable performance across the tasks.

Interactive Meaning Construction A Workshop at IWCS 2015 Proceedings

Robust Co-occurrence Quantification for Lexical Distributional Semantics

Previous optimisations of parameters affecting the word-context association measure used in distr... more Previous optimisations of parameters affecting the word-context association measure used in distributional vector space models have focused either on highdimensional vectors with hundreds of thousands of dimensions, or dense vectors with dimensionality of a few hundreds; but dimensionality of a few thousands is often applied in compositional tasks as it is still computationally feasible and does not require the dimensionality reduction step. We present a systematic study of the interaction of the parameters of the association measure and vector dimensionality, and derive parameter selection heuristics that achieve performance across word similarity and relevance datasets competitive with the results previously reported in the literature achieved by highly dimensional or dense models.

Alzheimer’s Dementia Recognition From Spontaneous Speech Using Disfluency and Interactional Features

Frontiers in computer science, Jun 18, 2021

Alzheimer's disease (AD) is a progressive, neurodegenerative disorder mainly characterized by mem... more Alzheimer's disease (AD) is a progressive, neurodegenerative disorder mainly characterized by memory loss with deficits in other cognitive domains, including language, visuospatial abilities, and changes in behavior. Detecting diagnostic biomarkers that are noninvasive and cost-effective is of great value not only for clinical assessments and diagnostics but also for research purposes. Several previous studies have investigated AD diagnosis via the acoustic, lexical, syntactic, and semantic aspects of speech and language. Other studies include approaches from conversation analysis that look at more interactional aspects, showing that disfluencies such as fillers and repairs, and purely nonverbal features such as inter-speaker silence, can be key features of AD conversations. These kinds of features, if useful for diagnosis, may have many advantages: They are simple to extract and relatively language-, topic-, and taskindependent. This study aims to quantify the role and contribution of these features of interaction structure in predicting whether a dialogue participant has AD. We used a subset of the Carolinas Conversation Collection dataset of patients with AD at moderate stage within the age range 60-89 and similar-aged non-AD patients with other health conditions. Our feature analysis comprised two sets: disfluency features, including indicators such as self-repairs and fillers, and interactional features, including overlaps, turn-taking behavior, and distributions of different types of silence both within patient speech and between patient and interviewer speech. Statistical analysis showed significant differences between AD and non-AD groups for several disfluency features (edit terms, verbatim repeats, and substitutions) and interactional features (lapses, gaps, attributable silences, turn switches per minute, standardized phonation time, and turn length). For the classification of AD patient conversations vs. non-AD patient conversations, we achieved 83% accuracy with disfluency features, 83% accuracy with interactional features, and an overall accuracy of 90% when combining both feature sets using support vector machine classifiers. The discriminative power of these features, perhaps combined with more conventional linguistic features, therefore shows potential for integration into noninvasive clinical assessments for AD at advanced stages.

On the Means for Clarification in Dialogue

Springer eBooks, 2003

The ability to request clarification of utterances is a vital part of the communicative process. ... more The ability to request clarification of utterances is a vital part of the communicative process. In this paper we discuss the range of possible forms for clarification requests, together with the range of readings they can convey. We present the results of corpus analysis which show a correlation between certain forms and possible readings, together with some indication of maximum likely distance between request and the utterance being clarified. We then explain the implications of these results for a possible HPSG analysis of clarification requests and for an ongoing implementation of a clarification-capable dialogue system.

The Web We Weave: Untangling the Social Graph of the IETF

Proceedings of the International AAAI Conference on Web and Social Media, May 31, 2022

The Internet Engineering Task Force (IETF) has developed many of the technical standards that und... more The Internet Engineering Task Force (IETF) has developed many of the technical standards that underpin the Internet. The standards development process followed by the IETF is open and consensus-driven, but is inherently both a social and political activity, and latent influential structures might exist within the community. Exploring and understanding these structures is essential to ensuring the IETF's resilience and openness. We use network analysis to explore the social graph of IETF participants, based on public email discussions and co-author relationships, and the influence of key contributors. We show that a small core of participants dominates: the top 10% contribute almost half (43.75%) of the emails and come from a relatively small group of organisations. On the other hand, we also find that influence has become relatively more decentralised with time. IETF participants also propose and work on drafts that are either adopted by a working group for further refinement or get rejected at an early stage. Using the social graph features combined with email text features, we perform regression analysis to understand the effect of user influence on the success of new work being adopted by the IETF. Our findings shed useful insights into the behavior of participants across time, correlation between influence and success in draft adoption, and the significance of affiliated organisations in the authorship of drafts.

Ludics, Dialogue and Interaction

The use of general descriptive names, registered names, trademarks, etc. in this publication does... more

Lattice Theoretic Relevance in Incremental Reference Processing

Incrementality, Alignment and Shared Utterances

Institute of Philosophy, 2004

This paper describes an implemented prototype dialogue model within the Dynamic Syntax (DS) frame... more This paper describes an implemented prototype dialogue model within the Dynamic Syntax (DS) framework which directly reflects dialogue phenomena such as alignment, routinization and shared utterances. In DS, word-by-word incremental parsing and generation are defined in terms of actions on semantic tree structures. This paper proposes a model of dialogue context which includes these trees and their associated actions, and shows how alignment and routinization result directly from minimisation of lexicon search (and hence speaker's effort), and how switch of speaker/hearer roles in shared utterances can be seen as a switch between incremental processes directed by different goals, but sharing the same (partial) data structures. 1 Example (1) from the BNC, file KNY (sentences 315-317). 2 Although see for an initial DRT-based approach.

Probabilistic Record Type Lattices for Incremental Reference Processing

Studies in linguistics and philosophy, 2017

Completability vs (In)completeness

Acta Linguistica Hafniensia, 2020

In natural conversation, no notion of "complete sentence" is required for syntactic licensing. Bu... more In natural conversation, no notion of "complete sentence" is required for syntactic licensing. But so-called "fragmentary", "incomplete", and abandoned utterances constitute problematic data for standard formalisms. We argue, instead, that such data contextualised show that: (a) non-sentential utterances are adequate to underpin people's coordination, while (b) all linguistic dependencies can be systematically distributed across more than one participant and turn. Standard models have problems accounting for such data because their notions of 'constituency' and 'syntactic domain' are independent of performance considerations. Concomitantly, no notion of "full proposition" or encoded speech act is necessary for successful interaction. Strings, contents, and joint actions emerge incrementally 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 online, at each step affording and constraining possibilities for further action. For this reason, they need to be modelled as such: a representational level of abstract syntax, divorced from conceptual structure and physical action, impedes a natural account of subsentential coordination phenomena. Instead, we argue that we need a view of grammar as a "skill" employing domain-general mechanisms rather than fixed form-meaning mappings. 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".

Information-Theoretic Segmentation of Natural Language

We present computational experiments on language segmentation using a general information-theoret... more We present computational experiments on language segmentation using a general information-theoretic cognitive model. We present a method which uses the statistical regularities of language to segment a continuous stream of symbols into “meaningful units” at a range of levels. Given a string of symbols—in the present approach, textual representations of phonemes—we attempt to find the syllables such as grea and sy (in the word greasy); words such as in, greasy, wash, and water ; and phrases such as in greasy wash water. The approach is entirely information-theoretic, and requires no knowledge of the units themselves; it is thus assumed to require only general cognitive abilities, and has previously been applied to music. We tested our approach on two spoken language corpora, and we discuss our results in the context of learning as a statistical processes.

Exploring Semantic Incrementality with Dynamic Syntax and Vector Space Semantics

ArXiv, 2018

One of the fundamental requirements for models of semantic processing in dialogue is incrementali... more One of the fundamental requirements for models of semantic processing in dialogue is incrementality: a model must reflect how people interpret and generate language at least on a word-by-word basis, and handle phenomena such as fragments, incomplete and jointly-produced utterances. We show that the incremental word-by-word parsing process of Dynamic Syntax (DS) can be assigned a compositional distributional semantics, with the composition operator of DS corresponding to the general operation of tensor contraction from multilinear algebra. We provide abstract semantic decorations for the nodes of DS trees, in terms of vectors, tensors, and sums thereof; using the latter to model the underspecified elements crucial to assigning partial representations during incremental processing. As a working example, we give an instantiation of this theory using plausibility tensors of compositional distributional semantics, and show how our framework can incrementally assign a semantic plausibilit...

Rare-Class Dialogue Act Tagging for Alzheimer’s Disease Diagnosis

Alzheimer’s Disease (AD) is associated with many characteristic changes, not only in an individua... more Alzheimer’s Disease (AD) is associated with many characteristic changes, not only in an individual’s language but also in the interactive patterns observed in dialogue. The most indicative changes of this latter kind tend to be associated with relatively rare dialogue acts (DAs), such as those involved in clarification exchanges and responses to particular kinds of questions. However, most existing work in DA tagging focuses on improving average performance, effectively prioritizing more frequent classes; it thus gives a poor performance on these rarer classes and is not suited for application to AD analysis. In this paper, we investigate tagging specifically for rare class DAs, using a hierarchical BiLSTM model with various ways of incorporating information from previous utterances and DA tags in context. We show that this can give good performance for rare DA classes on both the general Switchboard corpus (SwDA) and an AD-specific conversational dataset, the Carolinas Conversation...

Shifting Opinions: Experiments on Agreement and Disagreement in Dialogue

Disagreement is understood to be socially problematic; it also rarely surfaces in naturally occur... more Disagreement is understood to be socially problematic; it also rarely surfaces in naturally occurring conversation. An experiment was designed to allow us to directly manipulate the occurrence of exposed (dis)agreement and track its effects on the subsequent dialogue. This is the first experiment to directly manipulate the occurrence of exposed agreement and disagreement in dialogue. Insertions of exposed disagreement disrupt dialogues, bringing the topic of disagreement directly into the conversation, provoking clarification requests and resulting in a greater number of self-edits when formulating turns. The insertion of disagreement also led to more instances of exposed agreement, suggesting that dialogue partners co-operate to redress the face-threat of disagreement. Conversely, exposed agreement insertions were not as incongruous and had less disruptive impact on the ensuing dialogues; however, introducing agreement into the dialogue did lead to greater deliberation, with more a...