Section: New Results

Discourse Dynamics

Participants : Maxime Amblard, Timothée Bernard, Clément Beysson, Maria Boritchev, Philippe de Groote, Bruno Guillaume, Pierre Ludmann, Michel Musiol.

Dynamic Logic

We have revisited the type-theoretic dynamic logic introduced in [3]. We have shown how a slightly richer notion of continuation together with an appropriate notion of polarity results in a richer and more powerful framework. In particular, it allows new dynamic connectives and quantifiers to be defined in a systematic way. This work has been presented as an invited talk at the LACompLing 2018 symposium [11].

Discourse Relations

A text as a whole must exhibit some coherence that makes it more than just a bag of sentences. This coherence hinges on discourse relations (DRs), that express the articulations between the different segments of the text. Typical DRs include relations of Contrast, Consequence or Explanation. The most direct and reliable way to express a DR is to use a discourse connective (e.g., because, instead, for example). These lexical items have specific syntactic, semantic and pragmatic properties, the study of which is the subject of Timothée Bernard's PhD thesis.

Some discourse connectives (typically, adverbial connectives such as so or otherwise) have only one syntactic argument. It then seems natural to use an anaphora mechanism to retrieve the other argument from the context. This proposal has been formalized in [12] by means of continuation-based type theoretic dynamic logic. In this model, the semantic arguments of a DR are considered to be abstract entities akin to Davidsonian events. This approach raises difficulties when the argument of DR is a negative sentence. Indeed, according to the standard analysis of negation in event semantics, a negative sentence does not introduce any specific event. In order to circumvent this problem, we have developped a logical theory of negative events [13], [17], [29].

Dynamic Generalized Quantifiers

Clement Beysson has continued his work on dynamic generalized quantifiers as denotations of the (French) determiners. In this context, he has studied several issues raised by the modeling of plural determiners. In particular, the opposition between distributive and collective interpretations suggests that intrinsically dynamic plural determiners should introduce plural discourse referents that stand for collection of entities. In order to formalize this notion, he has studied several theories of plurality: mereology, plural logic, and second-order logic.

Dialogue Modeling

Maxime Amblard and Maria Boritchev develop a dynamic approach of dialogue modelling. One of the main difference between discourse and dialogue is the interactions between the speakers. To do so, they introduce a formal approach to compositional processing of questions and answers. They address dialogue lexicality issues starting from the formal definitions of so-called Düsseldorf Frame Semantics given in [51]. They introduce a view of dialogues as compositions of negotiation phases that can be studied separately one from another while linked by a common dialogue context (accessible to all participants of a dialogue). They apply Inquisitive Semantics [39] in that context.

Maxime Amblard and Maria Boritchev works on the categorisation of questions and answers and apply some machine learning approches for automatic classification. They present the architecture of the model, especially how to handle these phenomena with logical representations in [14]. Their view is to narrow the problem of identifying incomprehension in dialogue to the one of finding logical incoherences in speech act combinations as the one we found in the SLAM project (ongoing project of the Sémagramme team on interviews with schizophrenics). They also start to build a new corpus - DinG (Discourse in Dialogue) - based on record and transcript plays to the settlers of Catan board game.

Maxime Amblard also started a cooperation with CLASP, especially with Robin Cooper, Ellen Breitholtz and Chris Howes. They work on the synchronisation of the representation of dialogue modelling with the previous proposals and Type-Theoretic-Records (TTR) [41]. They apply the solution on extracts from two corpora where patients with schizophrenia are involved.

Pathological Discourse Modelling

Michel Musiol obtained a part-time delegation in the Semagramme team. This proximity makes possible to set up a more active dialogue on the issue of pathological discourse modeling. He has worked on the development of the possibility of testing his conjectures on the cognitive and psychopathological profile of the interlocutors, in addition to information provided by the model of ruptures and incongruities in pathological discourse. This methodological system makes it possible to discuss, or even evaluate, the heuristic potential of the computational models developed on the basis of empirical facts.

Moreover, the diagnostic tools used today by the professional community (clinical and psychiatric) are of limited expertise for the effective identification of the signs of the pathology for at least two reasons: on the one hand, they are much too imprecise on the side of the recognition of Language Impairment and Thought Disorder (no underlying linguistic and psycholinguistic theories); on the other hand, they do not take into account (either theoretically or technically) the discursive structure within which these disorders are expressed. The objective of this research program is therefore also to anticipate the development of diagnostic tools for the psychiatric and psychological community.

As part of the work carried out in the SLAM project, Maxime Amblard, Michel Musiol and Manuel Rebuschi (Archives Henri-Poincaré, Université de Lorraine) continue to work on modelling interactions with schizophrenic patients. The project has progressed on three different operational levels: building new resources, editing a volume (Springer) on the SLAM project in 2019 and improving the representation model.

An agreement is being deployed with the psychiatric hospital of Aix-en-Provence. The on-site staff administered a test protocol to the entire test group of 60 people. Transcripts are in progress, which will provide a significant amount of data to work on for the project. Thanks to the involvement of a medical staff, the recovery of new data appears well advanced. In the same perspective, contacts are being made with the Psychotherapeutic Centre in Nancy.

In addition, Maxime Amblard carried out a one-week international mobility at CLASP thanks to a mobility grant from the French Embassy in Sweden. Discussions were initiated with these colleagues for the development of projects using formal semantic models for the analysis of interaction with schizophrenic patients.