Stanley Peters | Stanford University (original) (raw)

Papers by Stanley Peters

Artificial Intelligence in Education, Apr 1, 2006

In designing and building tutorial dialogue systems it is important not only to understand the ta... more In designing and building tutorial dialogue systems it is important not only to understand the tactics employed by human tutors but also to understand how tutors decide when to use various tactics. We argue that these decisions are based not only on student problem-solving steps and the content of student utterances, but also on the meta-communicative information conveyed through spoken utterances (e.g., pauses, disfluencies, intonation). Since this information is often infrequent or unavailable in typed input, tutorial dialogue systems with speech interfaces have the potential to be more effective than those without. This paper gives an overview of the Spoken Conversational Tutor (SCoT) that we have built and describes how we are beginning to make use of spoken language information in SCoT. Specifically, we describe a study aimed at using meta-communicative information to gauge student uncertainty and respond accordingly. In this study, we identify linguistic devices used by human tutors when responding to utterances containing signals of uncertainty, integrate these response strategies into two versions of SCoT, and evaluate their relative effectiveness. Our main hypothesis-that tutors are more effective if they use these linguistic devices in response to student uncertainty-was not confirmed, but our secondary hypothesis-that tutors using these linguistic devices are more effective than tutors that do not use them-was supported by the results. 1 A joint activity is an activity where participants coordinate with one another to achieve both public and private goals (Clark, 1996). Moving a desk, playing a duet, and shaking hands are all examples of joint activities.

This paper discusses the problem of second-person pronoun resolution in dialogue: determining who... more This paper discusses the problem of second-person pronoun resolution in dialogue: determining who (if anyone) the word ‘you’ refers to. We motivate the task, and break it down into three distinct subtasks – distinguishing generic from deictic uses, distinguishing singular from plural uses, and determining individual reference. We then describe a dataset and series of supervised classification experiments, and show that various linguistic and non-linguistic features can be used to achieve overall accuracies of up to 78%.

The paper describes evaluation resources for concept-based, cross-lingual information retrieval i... more The paper describes evaluation resources for concept-based, cross-lingual information retrieval in the medical domain. All resources were constructed in the context of the MuchMore project and are freely available through the project website. Available resources include: a bilingual, parallel document collection of German and English medical scientific abstracts, a set of queries and corresponding relevance assessments, two manually disambiguated test sets for semantic annotation (sense disambiguation), two evaluation lists for German morphological decomposition of medical terms.

Interspeech 2008, 2008

We address the problem of identifying words and phrases that accurately capture, or contribute to... more We address the problem of identifying words and phrases that accurately capture, or contribute to, the semantic gist of decisions made in multi-party human-human meetings. We first describe our approach to modelling decision discussions in spoken meetings and then compare two approaches to extracting information from these discussions. The first one uses an opendomain semantic parser that identifies candidate phrases for decision summaries and then employs machine learning techniques to select from those candidate phrases. The second one uses categorical and sequential classifiers that exploit simple syntactic and semantic features to identify words and phrases relevant for decision summarization.

Semantics and Linguistic Theory, 2015

Research on reciprocals has uncovered a variety of semantic contributions that the reciprocal can... more Research on reciprocals has uncovered a variety of semantic contributions that the reciprocal can make, creating problems for proposals that the reciprocal unambiguously means something weak (e.g., Langendoen 1978). However, there is no real evidence that reciprocals are ambiguous, despite previous claims to the contrary (e.g., Fiengo and Lasnik 1973). First, we classify the apparently heterogeneous list of meanings proposed in previous research into a natural taxonomy, showing how they arise from a small stock of logical operations and predicates. Second, we exhibit a partial ordering of the various reciprocal meanings according to logical strength, which we make crucial use of in determining what reciprocals mean in each specific context where they appear. Third, we hypothesize that a reciprocal statement expresses the strongest candidate meaning that is consistent with known properties of the relation expressed by the scope of the reciprocal. This hypothesis is supported by analy...

Proceedings of the 13th international conference on Intelligent user interfaces, 2008

2008 IEEE Spoken Language Technology Workshop, 2008

The CALO Meeting Assistant provides for distributed meeting capture, annotation, automatic transc... more The CALO Meeting Assistant provides for distributed meeting capture, annotation, automatic transcription and semantic analysis of multiparty meetings, and is part of the larger CALO personal assistant system. This paper summarizes the CALO-MA architecture and its speech recognition and understanding components, which include real-time and offline speech transcription, dialog act segmentation and tagging, question-answer pair identification, action item recognition, decision extraction, and summarization.

IEEE International Conference on Advanced Learning Technologies, 2004. Proceedings.

Contextualizing learning in an intelligent tutoring system is difficult for many reasons. Goals s... more Contextualizing learning in an intelligent tutoring system is difficult for many reasons. Goals such as presenting material in an understandable manner, minimizing confusion and frustration, and helping the student reason about their actions all need to be balanced. Previous research has shown reflective discussions (with human tutors) occurring after problem-solving to be effective in helping students reason about their own actions [14]. However, leading a reflective discussion makes it difficult to present information in an understandable manner, and without contextualization it is easy to create student confusion and frustration. This raises the question: how can intelligent tutoring systems effectively contextualize learning in a reflective discussion? In this paper we describe the tutorial architecture of SCoT, a Spoken Conversational Tutor that uses flexible, adaptive planning and multi-modal task modeling to support the contextualization of learning in reflective dialogues.

Proceedings of the 12th Conference of the European Chapter of the Association for Computational Linguistics on - EACL '09, 2009

We explore the problem of resolving the second person English pronoun you in multi-party dialogue... more We explore the problem of resolving the second person English pronoun you in multi-party dialogue, using a combination of linguistic and visual features. First, we distinguish generic and referential uses, then we classify the referential uses as either plural or singular, and finally, for the latter cases, we identify the addressee. In our first set of experiments, the linguistic and visual features are derived from manual transcriptions and annotations, but in the second set, they are generated through entirely automatic means. Results show that a multimodal system is often preferable to a unimodal one. * We thank the anonymous EACL reviewers, and Surabhi Gupta, John Niekrasz and David Demirdjian for their comments and technical assistance. This work was supported by the CALO project (DARPA grant NBCH-D-03-0010). 1 See e.g.

Proceedings of the SIGDIAL 2009 Conference on The 10th Annual Meeting of the Special Interest Group on Discourse and Dialogue - SIGDIAL '09, 2009

This paper examines the resolution of the second person English pronoun you in multi-party dialog... more This paper examines the resolution of the second person English pronoun you in multi-party dialogue. Following previous work, we attempt to classify instances as generic or referential, and in the latter case identify the singular or plural addressee. We show that accuracy and robustness can be improved by use of simple lexical features, capturing the intuition that different uses and addressees are associated with different vocabularies; and we show that there is an advantage to treating referentiality and addressee identification as separate (but connected) problems.

Proceedings of the SIGDIAL 2009 Conference on The 10th Annual Meeting of the Special Interest Group on Discourse and Dialogue - SIGDIAL '09, 2009

We use directed graphical models (DGMs) to automatically detect decision discussions in multi-par... more We use directed graphical models (DGMs) to automatically detect decision discussions in multi-party dialogue. Our approach distinguishes between different dialogue act (DA) types based on their role in the formulation of a decision. DGMs enable us to model dependencies, including sequential ones. We summarize decisions by extracting suitable phrases from DAs that concern the issue under discussion and its resolution. Here we use a semantic-similarity metric to improve results on both manual and ASR transcripts.

The reciprocal construction also imposes a nonidentity requirement between the filler of the posi... more The reciprocal construction also imposes a nonidentity requirement between the filler of the position linked to the antecedent of the reciprocal and the filler of a position we will refer to as the reciprocalized argument. Heim, Lasnik, and May (1991) call this requirement reciprocation. In (1) the antecedent of the reciprocal is linked to the agent argument position of hit, and the reciprocalized argument position is linked to the reciprocal phrase each other. The intended interpretation involves different individuals filling the agent and patient ...

This is is an experimental study of the semantics of the construction NP was (not) Adj to VP wher... more This is is an experimental study of the semantics of the construction NP was (not) Adj to VP where Adj is an evaluative adjective such as stupid. We show that in the simple past tense this construction is predominantly for most people but for some. We also demonstrate that the interpretations are sensitive to preconceptions about how suitable the adjective is as a characterization of the event described by the innitival clause. This C /D eect gives the construction its chameleon-like characteristics.

Proceedings of Human Language Technologies: The Annual Conference of the North American Chapter of the Association for Computational Linguistics: Demonstrations on XX - NAACL '07, 2007

In this demonstration we present a conversational dialog system for automobile drivers. The syste... more In this demonstration we present a conversational dialog system for automobile drivers. The system provides a voicebased interface to playing music, finding restaurants, and navigating while driving. The design of the system as well as the new technologies developed will be presented. Our evaluation showed that the system is promising, achieving high task completion rate and good user satisfation.

Studies in Linguistics and Philosophy, 1982

Chomsky's argument that natural languages are not finite state languages puts a lowe... more Chomsky's argument that natural languages are not finite state languages puts a lower bound on the weak generative capacity of grammars for natural languages (Chomsky (1956)). Arguments based on weak generative capacity are useful in excluding classes of formal devices as characterizations of natural language, but they are not the only formal considerations by which this can be done. Generative grammars may also be excluded because they cannot assign the correct structural descriptions to the terminal strings of a ...

Proceedings of the 9th SIGdial Workshop on Discourse and Dialogue - SIGdial '08, 2008

We describe a process for automatically detecting decision-making sub-dialogues in transcripts of... more We describe a process for automatically detecting decision-making sub-dialogues in transcripts of multi-party, human-human meetings. Extending our previous work on action item identification, we propose a structured approach that takes into account the different roles utterances play in the decisionmaking process. We show that this structured approach outperforms the accuracy achieved by existing decision detection systems based on flat annotations, while enabling the extraction of more fine-grained information that can be used for summarization and reporting.

Lecture Notes in Computer Science, 2004

IEEE Transactions on Audio, Speech, and Language Processing, 2010

The CALO Meeting Assistant (MA) provides for distributed meeting capture, annotation, automatic t... more The CALO Meeting Assistant (MA) provides for distributed meeting capture, annotation, automatic transcription and semantic analysis of multiparty meetings, and is part of the larger CALO personal assistant system. This paper presents the CALO-MA architecture and its speech recognition and understanding components, which include real-time and offline speech transcription, dialog act segmentation and tagging, topic identification and segmentation, question-answer pair identification, action item recognition, decision extraction, and summarization.

csli-publications.stanford.edu

Various challenges have emerged over several years of grammar engineering for the spoken dialogue... more Various challenges have emerged over several years of grammar engineering for the spoken dialogue interface to the Navy damage control simulator DC-Train and the Spoken Conversational Tutor SCoT-DC, which reviews DC-Train performance. The ...

staff.science.uva.nl

In face-to-face meetings, assigning and agreeing to carry out future actions is a frequent subjec... more In face-to-face meetings, assigning and agreeing to carry out future actions is a frequent subject of conversation. Work thus far on identifying these action item discussions has focused on extracting them from entire transcripts of meetings. Here we investigate a human-initiative targeting approach by simulating a scenario where meeting participants provide low-load input (pressing a button during the dialogue) to indicate that an action item is being discussed. We compare the performance of categorical and sequential machine learning methods and their robustness when the point of user input varies. We also consider automatic summarization of action items in cases where individual utterances contain more than one type of relevant information.

Artificial Intelligence in Education, Apr 1, 2006

In designing and building tutorial dialogue systems it is important not only to understand the ta... more In designing and building tutorial dialogue systems it is important not only to understand the tactics employed by human tutors but also to understand how tutors decide when to use various tactics. We argue that these decisions are based not only on student problem-solving steps and the content of student utterances, but also on the meta-communicative information conveyed through spoken utterances (e.g., pauses, disfluencies, intonation). Since this information is often infrequent or unavailable in typed input, tutorial dialogue systems with speech interfaces have the potential to be more effective than those without. This paper gives an overview of the Spoken Conversational Tutor (SCoT) that we have built and describes how we are beginning to make use of spoken language information in SCoT. Specifically, we describe a study aimed at using meta-communicative information to gauge student uncertainty and respond accordingly. In this study, we identify linguistic devices used by human tutors when responding to utterances containing signals of uncertainty, integrate these response strategies into two versions of SCoT, and evaluate their relative effectiveness. Our main hypothesis-that tutors are more effective if they use these linguistic devices in response to student uncertainty-was not confirmed, but our secondary hypothesis-that tutors using these linguistic devices are more effective than tutors that do not use them-was supported by the results. 1 A joint activity is an activity where participants coordinate with one another to achieve both public and private goals (Clark, 1996). Moving a desk, playing a duet, and shaking hands are all examples of joint activities.

This paper discusses the problem of second-person pronoun resolution in dialogue: determining who... more This paper discusses the problem of second-person pronoun resolution in dialogue: determining who (if anyone) the word ‘you’ refers to. We motivate the task, and break it down into three distinct subtasks – distinguishing generic from deictic uses, distinguishing singular from plural uses, and determining individual reference. We then describe a dataset and series of supervised classification experiments, and show that various linguistic and non-linguistic features can be used to achieve overall accuracies of up to 78%.

The paper describes evaluation resources for concept-based, cross-lingual information retrieval i... more The paper describes evaluation resources for concept-based, cross-lingual information retrieval in the medical domain. All resources were constructed in the context of the MuchMore project and are freely available through the project website. Available resources include: a bilingual, parallel document collection of German and English medical scientific abstracts, a set of queries and corresponding relevance assessments, two manually disambiguated test sets for semantic annotation (sense disambiguation), two evaluation lists for German morphological decomposition of medical terms.

Interspeech 2008, 2008

We address the problem of identifying words and phrases that accurately capture, or contribute to... more We address the problem of identifying words and phrases that accurately capture, or contribute to, the semantic gist of decisions made in multi-party human-human meetings. We first describe our approach to modelling decision discussions in spoken meetings and then compare two approaches to extracting information from these discussions. The first one uses an opendomain semantic parser that identifies candidate phrases for decision summaries and then employs machine learning techniques to select from those candidate phrases. The second one uses categorical and sequential classifiers that exploit simple syntactic and semantic features to identify words and phrases relevant for decision summarization.

Semantics and Linguistic Theory, 2015

Research on reciprocals has uncovered a variety of semantic contributions that the reciprocal can... more Research on reciprocals has uncovered a variety of semantic contributions that the reciprocal can make, creating problems for proposals that the reciprocal unambiguously means something weak (e.g., Langendoen 1978). However, there is no real evidence that reciprocals are ambiguous, despite previous claims to the contrary (e.g., Fiengo and Lasnik 1973). First, we classify the apparently heterogeneous list of meanings proposed in previous research into a natural taxonomy, showing how they arise from a small stock of logical operations and predicates. Second, we exhibit a partial ordering of the various reciprocal meanings according to logical strength, which we make crucial use of in determining what reciprocals mean in each specific context where they appear. Third, we hypothesize that a reciprocal statement expresses the strongest candidate meaning that is consistent with known properties of the relation expressed by the scope of the reciprocal. This hypothesis is supported by analy...

Proceedings of the 13th international conference on Intelligent user interfaces, 2008

2008 IEEE Spoken Language Technology Workshop, 2008

The CALO Meeting Assistant provides for distributed meeting capture, annotation, automatic transc... more The CALO Meeting Assistant provides for distributed meeting capture, annotation, automatic transcription and semantic analysis of multiparty meetings, and is part of the larger CALO personal assistant system. This paper summarizes the CALO-MA architecture and its speech recognition and understanding components, which include real-time and offline speech transcription, dialog act segmentation and tagging, question-answer pair identification, action item recognition, decision extraction, and summarization.

IEEE International Conference on Advanced Learning Technologies, 2004. Proceedings.

Contextualizing learning in an intelligent tutoring system is difficult for many reasons. Goals s... more Contextualizing learning in an intelligent tutoring system is difficult for many reasons. Goals such as presenting material in an understandable manner, minimizing confusion and frustration, and helping the student reason about their actions all need to be balanced. Previous research has shown reflective discussions (with human tutors) occurring after problem-solving to be effective in helping students reason about their own actions [14]. However, leading a reflective discussion makes it difficult to present information in an understandable manner, and without contextualization it is easy to create student confusion and frustration. This raises the question: how can intelligent tutoring systems effectively contextualize learning in a reflective discussion? In this paper we describe the tutorial architecture of SCoT, a Spoken Conversational Tutor that uses flexible, adaptive planning and multi-modal task modeling to support the contextualization of learning in reflective dialogues.

Proceedings of the 12th Conference of the European Chapter of the Association for Computational Linguistics on - EACL '09, 2009

We explore the problem of resolving the second person English pronoun you in multi-party dialogue... more We explore the problem of resolving the second person English pronoun you in multi-party dialogue, using a combination of linguistic and visual features. First, we distinguish generic and referential uses, then we classify the referential uses as either plural or singular, and finally, for the latter cases, we identify the addressee. In our first set of experiments, the linguistic and visual features are derived from manual transcriptions and annotations, but in the second set, they are generated through entirely automatic means. Results show that a multimodal system is often preferable to a unimodal one. * We thank the anonymous EACL reviewers, and Surabhi Gupta, John Niekrasz and David Demirdjian for their comments and technical assistance. This work was supported by the CALO project (DARPA grant NBCH-D-03-0010). 1 See e.g.

Proceedings of the SIGDIAL 2009 Conference on The 10th Annual Meeting of the Special Interest Group on Discourse and Dialogue - SIGDIAL '09, 2009

This paper examines the resolution of the second person English pronoun you in multi-party dialog... more This paper examines the resolution of the second person English pronoun you in multi-party dialogue. Following previous work, we attempt to classify instances as generic or referential, and in the latter case identify the singular or plural addressee. We show that accuracy and robustness can be improved by use of simple lexical features, capturing the intuition that different uses and addressees are associated with different vocabularies; and we show that there is an advantage to treating referentiality and addressee identification as separate (but connected) problems.

Proceedings of the SIGDIAL 2009 Conference on The 10th Annual Meeting of the Special Interest Group on Discourse and Dialogue - SIGDIAL '09, 2009

We use directed graphical models (DGMs) to automatically detect decision discussions in multi-par... more We use directed graphical models (DGMs) to automatically detect decision discussions in multi-party dialogue. Our approach distinguishes between different dialogue act (DA) types based on their role in the formulation of a decision. DGMs enable us to model dependencies, including sequential ones. We summarize decisions by extracting suitable phrases from DAs that concern the issue under discussion and its resolution. Here we use a semantic-similarity metric to improve results on both manual and ASR transcripts.

The reciprocal construction also imposes a nonidentity requirement between the filler of the posi... more The reciprocal construction also imposes a nonidentity requirement between the filler of the position linked to the antecedent of the reciprocal and the filler of a position we will refer to as the reciprocalized argument. Heim, Lasnik, and May (1991) call this requirement reciprocation. In (1) the antecedent of the reciprocal is linked to the agent argument position of hit, and the reciprocalized argument position is linked to the reciprocal phrase each other. The intended interpretation involves different individuals filling the agent and patient ...

This is is an experimental study of the semantics of the construction NP was (not) Adj to VP wher... more This is is an experimental study of the semantics of the construction NP was (not) Adj to VP where Adj is an evaluative adjective such as stupid. We show that in the simple past tense this construction is predominantly for most people but for some. We also demonstrate that the interpretations are sensitive to preconceptions about how suitable the adjective is as a characterization of the event described by the innitival clause. This C /D eect gives the construction its chameleon-like characteristics.

Proceedings of Human Language Technologies: The Annual Conference of the North American Chapter of the Association for Computational Linguistics: Demonstrations on XX - NAACL '07, 2007

In this demonstration we present a conversational dialog system for automobile drivers. The syste... more In this demonstration we present a conversational dialog system for automobile drivers. The system provides a voicebased interface to playing music, finding restaurants, and navigating while driving. The design of the system as well as the new technologies developed will be presented. Our evaluation showed that the system is promising, achieving high task completion rate and good user satisfation.

Studies in Linguistics and Philosophy, 1982

Chomsky's argument that natural languages are not finite state languages puts a lowe... more Chomsky's argument that natural languages are not finite state languages puts a lower bound on the weak generative capacity of grammars for natural languages (Chomsky (1956)). Arguments based on weak generative capacity are useful in excluding classes of formal devices as characterizations of natural language, but they are not the only formal considerations by which this can be done. Generative grammars may also be excluded because they cannot assign the correct structural descriptions to the terminal strings of a ...

Proceedings of the 9th SIGdial Workshop on Discourse and Dialogue - SIGdial '08, 2008

We describe a process for automatically detecting decision-making sub-dialogues in transcripts of... more We describe a process for automatically detecting decision-making sub-dialogues in transcripts of multi-party, human-human meetings. Extending our previous work on action item identification, we propose a structured approach that takes into account the different roles utterances play in the decisionmaking process. We show that this structured approach outperforms the accuracy achieved by existing decision detection systems based on flat annotations, while enabling the extraction of more fine-grained information that can be used for summarization and reporting.

Lecture Notes in Computer Science, 2004

IEEE Transactions on Audio, Speech, and Language Processing, 2010

The CALO Meeting Assistant (MA) provides for distributed meeting capture, annotation, automatic t... more The CALO Meeting Assistant (MA) provides for distributed meeting capture, annotation, automatic transcription and semantic analysis of multiparty meetings, and is part of the larger CALO personal assistant system. This paper presents the CALO-MA architecture and its speech recognition and understanding components, which include real-time and offline speech transcription, dialog act segmentation and tagging, topic identification and segmentation, question-answer pair identification, action item recognition, decision extraction, and summarization.

csli-publications.stanford.edu

Various challenges have emerged over several years of grammar engineering for the spoken dialogue... more Various challenges have emerged over several years of grammar engineering for the spoken dialogue interface to the Navy damage control simulator DC-Train and the Spoken Conversational Tutor SCoT-DC, which reviews DC-Train performance. The ...

staff.science.uva.nl

In face-to-face meetings, assigning and agreeing to carry out future actions is a frequent subjec... more In face-to-face meetings, assigning and agreeing to carry out future actions is a frequent subject of conversation. Work thus far on identifying these action item discussions has focused on extracting them from entire transcripts of meetings. Here we investigate a human-initiative targeting approach by simulating a scenario where meeting participants provide low-load input (pressing a button during the dialogue) to indicate that an action item is being discussed. We compare the performance of categorical and sequential machine learning methods and their robustness when the point of user input varies. We also consider automatic summarization of action items in cases where individual utterances contain more than one type of relevant information.