You can find here bibliographic citations (with abstracts) for most of my papers, divided by category and in inverted chronological order (most recent papers first).
2006
Tesar, R., V. Stmad, K. Jezek, and M. Poesio, 2006. Extending the single words-based document model: a comparison of bigrams and 2-itemsets. Proc. of ACM Symposium on Document Engineering, Amsterdam October 10-13, p. 138-146. (pdf)
Sutcliffe, R. F. E., J. Steinberger, U. Kruschwitz, M. Kabadjov and M. Poesio, 2006. Identifying Novel Information using Latent Semantic Analysis in the WIQA task at CLEF 2006. Proc. of CLEF, Alicante, September 20-22.
Almuhareb, A. and M. Poesio, 2006. MSDA: A Word Sense Discrimination Algorithm. Proc. of ECAI, Riva del Garda, August. (pdf)
Poesio, M., M. A. Kabadjov, P. Goux, L. Corti and E. Bishop, 2006. An Anonymization Module Based on Anaphora Resolution. Proc. of LREC, Genoa, May. (pdf)
Sanchez-Graillet, O., M. Poesio, M. Kabadjov and R. Tesar, 2006. What kind of problems do protein interactions raise for anaphora resolution? - A preliminary analysis. Proc of SMBM, Jena, April 9-12. (pdf)
2005
Steinberger, J., M. A. Kabadjov, and M. Poesio, 2005. Improving LSA-based Summarization with Anaphora Resolution. Proc. of HLT / EMNLP, Vancouver, October. (pdf)
Kabadjov, M. A., M. Poesio, and J. Steinberger, 2005. Task-Based Evaluation of Anaphora Resolution: The Case of Summarization. Proc. of RANLP Workshop on Recent Developments in Summarization, Varna, Bulgaria, September. (pdf)
Filik, R., A. J. Sanford, P. Sturt, and M. Poesio, 2005. Underspecification in anaphoric reference to structured entities. Proc. of AMLAP, (Full presentation), Ghent, September. (Pdf of abstract) (pdf of slides)
Almuhareb, A., D. Vinson, M. Poesio and G. Vigliocco, 2005. Speaker-generated and web-generated features: a comparison. Proc. of AMLAP, (Poster), Ghent, September. (pdf)
Artstein, R. and M. Poesio, 2005. Bias Decreases in Proportion to the Number of Annotators. Proc. of Workshop on Formal Grammar / the Mathematics of Language, Edinburgh, August, p. 141-150. (pdf)
Almuhareb, A. and M. Poesio, 2005. Concept Learning and Categorization from the Web. Proc. of Annual Meeting of the Cognitive Science Society, (Poster), Stresa, July. (pdf)
Almuhareb, A. and M. Poesio, 2005. Finding Concept Attributes in the Web. Proc. of the Corpus Linguistics Conference, Birmingham, July. (pdf)
Poesio, M. and R. Artstein, 2005. Annotating (Anaphoric) Ambiguity. Proc. of the Corpus Linguistics Conference, Birmingham, July. (pdf)
Pustejovsky, J., A. Meyers, M. Palmer and M. Poesio, 2005. Merging PropBank, NomBank, TimeBank, Penn Discourse Treebank and Coreference. Proc. of ACL Workshop on Frontiers in Corpus Annotation, Ann Arbor, Michigan, June.
Poesio, M. and R. Artstein, 2005. The Reliability of Anaphoric Annotation, Reconsidered: Taking Ambiguity into Account. Proc. of ACL Workshop on Frontiers in Corpus Annotation, Ann Arbor, Michigan, June, p. 76-83. (pdf)
Poesio, M. and A. Almuhareb, 2005. Identifying Concept Attributes Using a Classifier. In T. Baldwin, A. Korhonen and A. Villavicencio (eds), Proc. of ACL Workshop on Deep Lexical Semantics, Ann Arbor, Michigan, June. (pdf)
Poesio, M., M. Alexandrov-Kabadjov, R. Vieira, R. Goulart, and O. Uryupina, 2005. Does discourse-new detection help definite description resolution? Proc. of the Sixth IWCS, Tilburg, January. (pdf)
2004
2003
2002
2001
2000
We are annotating a corpus with information relevant to discourse entity realization, and especially the information needed to decide which type of NP to use. The corpus is being used to study correlations between NP type and certain semantic or discourse features, to evaluate hand-coded algorithms, and to train statistical models. We report on the development of our annotation scheme, the problems we have encountered, and the results obtained so far.
This paper describes an implementation of some key aspects of a theory of dialogue processing whose main concerns are to provide models of GROUNDING and of the role of DISCOURSE OBLIGATIONS in an agent's deliberation processes. Our system uses the TrindiKit dialogue move engine toolkit, which assumes a model of dialogue in which a participant's knowledge is characterised in terms of INFORMATION STATES which are subject to various kinds of updating mechanisms.
We discuss the results of an analysis of a corpus of spoken dialogues which led us to identify a few types of situations in which it appears that referring expressions cannot be completely resolved, and propose a generalization concerning the conditions under which this may be possible.
Older papers (before 2000)
We look at the effect of using high level discourse knowledge in dialogue act type detection. We also look at ways this knowledge can be used for improving language modelling and intonation modelling of utterance types. We find a significant improvement of predictability of dialogue models using higher level discourse knowledge.
We report preliminary results on training a statistical model of the task of {\NP} type determination. We explain our characterization of the task, discuss how our corpus was annotated, and present our preliminary results, discussing some problems raised by this work.
We report on the current proposal concerning the type of `coreference' annotation to be supported by the MATE workbench, motivating our proposal in relation to previous proposals in this area.
We are studying the feasibility of annotating a corpus with information relevant to NP generation - specifically, the information needed to decide which type of NP to use. Such a corpus might be used just to study correlations between NP type and certain semantic or discourse features, or to train statistical models. We report on the development of our annotation scheme, the problems we have encountered, and the results obtained so far.
Recognizing the dialogue act(s) performed by means of an utterance involves combining top-down expectations about the next likely `move' in a dialogue with bottom-up information extracted from the speech signal. We compared two ways of generating expectations: one which makes the expectations depend only on the previous act (as in a bigram model), and one which also takes into account the fact that individual dialogue acts play a role as part of larger conversational structures (`games'). Our models were built by training over the HCRC MapTask corpus using the LTG implementation of maximum entropy estimation. We achieved an accuracy of 38.6% using bigrams, of 50.6% taking game structure into account; adding information about speaker change resulted in an accuracy of 41.8% with bigrams, 54% with game structure. These results indicate that exploiting game structure does lead to improved expectations.
This is a preliminary study into the feasibility of constructing a spoken dialogue system for use in the teaching of English as a foreign language. Using the CSLU Rapid Prototyper (CSLUrp), we built two simple gamestyle CALL systems and tested them for usability with visiting students of English. Results are presented on various aspects of their functionality, and problems highlighted. Overall, it was found that usable CALL systems can be developed with current spoken dialogue technology.
Conversations involve all sorts of verbal activities beyond those strictly related to the performance of the task at hand. Among other things, the participants in a conversation have to make sure they both understand what's going on, to manage turn taking, and to keep each other informed about their progress in achieving their task. The participants share information about the status of all of these processes; this suggests that the view of the conversational score they share is rather more complex than assumed in previous accounts. We proposed a preliminary formalization of this more complex view of the conversational score in previous work; in this paper we revise that earlier model, and use our theory of the conversational score to give a partial specification of the effect of the dialogue acts included in the DRI classification.
We present preliminary results concerning the use of lexical clustering algorithms to acquire the kind of lexical knowledge needed to resolve definite descriptions, and in particular what we call `inferential' descriptions. We tested the hypothesis that the antecedent of an inferential description is primarily identified on the basis of its semantic distance from the description; we also tested several variants of the clustering algorithm. We found that the choice of parameters has a clear effect, and that the best results are obtained by measuring the distance between lexical vectors using the cosine measure. We also found, however, that factors other than semantic distance play the main role in the majority of cases; but in those cases in which the sort of lexical knowledge we acquired is the main factor, the algorithms we used performed reasonably well; several standing problems are discussed.
The effects of utterances such as cue phrases, keep-turn markers, and grounding signals cannot be characterized as changes to a shared record of the propositions under discussed: the simplest (and arguably most natural) way of characterizing the meaning of these utterances is in terms of a theory in which the conversational score is seen as a record of the discourse situation, or at least of the speech acts that have been performed. The problem then becomes to explain how discourse entities are accessible. We consider three hypotheses about the dynamics of a speech act-based theory of the conversational score, and argue that they could be implemented with relatively minor modifications to the technical tools already introduced in theories such as Compositional DRT.
Our goal is to develop a system capable of treating the largest possible subset of definite descriptions in unrestricted written texts. A previous prototype resolved anaphoric uses of definite descriptions and identified some types of first-mention uses, achieving a recall of 56%. In this paper we present the latest version of our system, which handles some types of bridging references, uses WordNet as a source of lexical knowledge, and achieves a recall of 65%.
All work on underspecification I am aware of is concerned with the computational and logical properties of underspecified representations, rather than with their role within a theory of grammar and a cognitively plausible theory of ambiguity processing. My goal in this paper is to flesh out the view of utterance processing underlying the `underspecification hypothesis', and to establish a connection between this perspective on utterance processing and current linguistic theory and psychological results.
It is argued that corpora should be taken as an opportunity to revise and improve our semantic theories, rather than a reason to switch to ad-hoc implementations. Two aspects of current semantic theory that do need work are dealing with ambiguity and with noisy input; some suggestions are made as to how to overcome these problems.
Underspecified Representations have been proposed as a way of formalizing sentence disambiguation in terms of (usually, nonmonotonic) reasoning. But in fact the existing theories about the semantics of underspecified representations either fail to capture our intuitions about ambiguity, or lead one to believe that the kind of inferences used in disambiguation have nothing to do with what we usually mean by inference. As a solution, I propose a theory of underspecification based on two hypotheses: that disambiguation is the task of recovering the content of the utterance events that take place in a discourse situation; and every utterance event that took place in a conversation, including utterances of single lexical items, is recorded in the common ground. The underspecified representations I propose are representations of what took place in the discourse situation, rather than directly about the content of an utterance. It is shown that the resulting theory of underspecification offers a more satisfactory account of lexical disambiguation and is more general that existing theories, in that can also be used to account for structural disambiguation.
We describe the goals, architecture, and functioning of the TRAINS-93 system, with emphasis on the representational issues involved in putting together a complex language processing and reasoning agent. The system is intended as an experimental prototype of an intelligent, conversationally proficient planning advisor in a dynamic domain of cargo trains and factories. For this team effort, our strategy at the outset was to let the designers of the various language processing, discourse processing, plan reasoning, execution and monitoring modules choose whatever representations seemed best suited for their tasks, but with the constraint that all should strive for principled, general approaches. Disparities between modules were bridged by careful design of the interfaces, based on regular in-depth discussion of issues encountered by the participants. Because of the goal of generality and principled representation, the multiple representations ended up with a good deal in common (for instance, the use of explicit event variables and the ability to refer to complex abstract objects such as plans); and future unifications seem quite possible. We explain some of the goals and particulars of the KRs used, evaluate the extent to which they served their purposes, and point out some of the tensions between representations that needed to be resolved. On the whole, we found that using very expressive representations minimized the tensions, since it is easier to extract what one needs from an elaborate representation retaining all semantic nuances, than to make up for lost information.
I present a theory of discourse interpretation based on the hypothesis that the common ground of a conversation contains a record not only of complete speech acts, but, more in general, of each action of uttering a contribution to the conversation: single words, word fragments, and fillers. I call the action of uttering a `minimal' contribution a micro conversational event. This model can serve as the basis for accounts of reference resolution in spoken conversations, as well as the interaction between parsing, repair, and reference resolution.
A formal analysis of ambiguity processing is a necessary prerequisite for the development of a theory of underspecification and discourse interpretation for {\NLP} systems. The analysis presented here is based on a distinction between semantic ambiguity and perceived ambiguity. A sentence is semantically ambiguous if it has a multiplicity of interpretations; a form of underspecified representation is introduced that can be used as the translation of a sentence that is semantically ambiguous in this sense. Perceived ambiguity, on the other hand, is captured in terms of hypothesis generation in context.
The long term goal of the TRAINS project is to develop an intelligent planning assistant that is conversationally proficient in natural language. The TRAINS system helps a user construct and monitor plans about a railroad freight system; their interaction takes place in natural language. The representational needs of a system like TRAINS include representing lexical meaning, dealing with the problem of ambiguity , make use of information about context, and finding a connection between the content of the current utterance and the plan being jointly developed by system and user . The goal of the paper is to describe how TRAINS-93, the latest prototype of TRAINS, deals with these issues.
The model of definite description interpretation in [Poesio, 1993] is revised to avoid that theory's assumption that the interpretation of definites occurs after scope interpretation has taken place. The processes involved in the interpretation of a definite are formalized as DRS construction rules applying in parallel to generate alternative hypotheses out of underspecified representations. Interpreting a definite requires identifying the particular situation, or discourse topic, that includes an object of that type. `Focus' effects are formulated in terms of situations and their hierarchical organization.
In this paper I discuss a class of definite descriptions that I will call Weak Definites. The data I will examine are exemplified by following sentences:(1) John got these data from the student of a linguist.
(2) I usually had breakfast at the corner of a major intersection.
What's interesting about definite descriptions of the form [NP the student of [NP a linguist]] is that they have a reading that is not predicted by either the Russellian theory of definite descriptions or Heim's theory. The reading of (1a) I am interested in can be paraphrased as: there is a linguist, and there is a student of that linguist, such that John got the data from that student. I propose that weak definites have this reading because their interpretation can be `anchored' to the interpretation of the NP that serves as complement of "of", so that if that NP gets bound by an operator, the NP gets bound as well.
This paper describes the TRAINS-93 Conversation System, an implemented system that acts as an intelligent planning assistant and converses with the user in natural language. The architecture of the system is described and particular attention is paid to the interactions between the language understanding and plan reasoning components. We examine how these two tasks constrain and inform each other in an integrated NL-based system.
I propose a situation-theoretic formalization of the organization of information in the common ground which makes it possible to formulate principles about (i) the way an interpretation for definite descriptions is chosen, and (ii) how the information used for interpreting definite descriptions changes during the conversation.
Reasoning about one's personal schedule of appointments is a common but surprisingly complex activity. Motivated by the novel application of planning and temporal reasoning techniques to this problem, we have extended the formalization of the temporal distance model of Dechter, Meiri, and Pearl. We have developed methods for using dates as reference intervals and for meeting the challenge of repeated activities, such as weekly recurring appointments.
The problem of ambiguity is central to any theory of language interpretation, whether we are interested in language processing in humans or in developing a usable natural language processing system. Psycholinguistic evidence suggests that human subjects are able to choose an interpretation when necessary, and that competing factors are involved in this choice; however, no theory of language interpretation deals satisfactorily with the combinatorial explosion puzzle---the fact that no matter how ambiguous natural language sentences are, they are usually interpreted without significant effort.The main idea presented in this dissertation is that the scope preferences observed in the literature are not the result of an independent `scope disambiguation' module, but of independent interpretation processes such as definite description interpretation or the interpretation of modals. None of these interpretive procedures is especially concerned with `scope disambiguation,' but the result of these inferences is that relations of contextual dependency such as anaphoric reference or presuppositionality become part of the common ground; the scope preferences observed in the literature reflect these relations of dependency. The dissertation includes a formal proposal concerning the representation of contextual dependency, and its impact on the semantics of sentence constituents.
My theory of ambiguity is based on a distinction between semantic ambiguity, that can be captured implicitly, by means of underspecified representations, and perceived ambiguity, that results from the process of discourse interpretation. My model of the common ground can be used to characterize both situations characterized by the presence of semantic ambiguity, and situations characterized by the existence of perceived ambiguity.
The reasoning that leads to the establishment of scoping preferences makes use, I argue, of information that is pragmatic in nature; this calls for a model of discourse interpretation in which the `common ground' contains such information. In the case of spoken language conversations, the common ground must be a model of the discourse situation of the conversational participants.