2023Activity reportProject-TeamSEMAGRAMME

ANR InExtenso:

WP4 of the project is dedicated to the evaluation of the environmental impact of the LLMs. More precisely, it aims at proposing a method for measuring the environmental impact of digital health and use it in the project evaluations and beyond.

7.1.1 ACGtk

• Name:
  Abstract Categorial Grammar Development Toolkit
• Keywords:
  Natural language processing, NLP, Syntactic analysis, Semantics
• Scientific Description:

  Abstract Categorial Grammars (ACG) are a grammatical formalism in which grammars are based on typed lambda-calculus. A grammar generates two languages: the abstract language (the language of parse structures), and the object language (the language of the surface forms, e.g., strings, or higher-order logical formulas), which is the realization of the abstract language.

  ACGtk provides two software tools to develop and to use ACGs: acgc, which is a grammar compiler, and acg, which is an interpreter of a command language that allows one, in particular, to parse and realize terms.

• Functional Description:
  ACGtk provides a piece of software for developing and using Abstract Categorial Grammars (ACG).
• Release Contributions:
  New version of the software that provides a new command language and new functionalities for the interpreter (completion, sorting of parsing structures, use of magic set rewriting technic for parsing). The compiler now accepts UTF-8, and a new syntax for identifiers, binders and operators was defined.
• URL:
  https://­gitlab.­inria.­fr/­ACG/­dev/­ACGtk
• Publications:
  hal-01242154, hal-01328702, tel-01412765, inria-00112956, inria-00100529
• Contact:
  Sylvain Pogodalla
• Participants:
  Philippe De Groote, Pierre Ludmann, Jiri Marsik, Sylvain Pogodalla, Vincent Tourneur

7.1.2 Grew

• Name:
  Graph Rewriting
• Keywords:
  Semantics, Syntactic analysis, NLP, Graph rewriting
• Functional Description:
  Grew is a Graph Rewriting tool dedicated to applications in NLP. Grew takes into account confluent and non-confluent graph rewriting and it includes several mechanisms that help to use graph rewriting in the context of NLP applications (built-in notion of feature structures, parametrization of rules with lexical information).
• News of the Year:
  In 2023, 4 new versions were released. The main new features are: non-injective matching, a new "with" keyword for positive filtering, a syntax extension to express disjunction on node matching.
• URL:
  https://­grew.­fr/
• Publications:
  hal-03724068, hal-03177701, hal-01930591, hal-01814386, hal-03021720, hal-04387830, hal-04387852, hal-03724068, hal-03724129, hal-03846825
• Contact:
  Bruno Guillaume
• Participants:
  Bruno Guillaume, Guy Perrier, Guillaume Bonfante

7.1.3 SLODiM

• Name:
  SLODiM
• Keywords:
  NLP, Discourse, Dialogue, French
• Functional Description:
  SLODiM is a software package for the analysis of oral French. It is more particularly developed to allow the analysis of interviews with clinicians in order to identify language behaviours characteristic of mental pathologies.
• Release Contributions:

  The latest version integrates new treatments in particular at the level of the identification of the backchannels.

  A version without the graphical representations is available without an account. Its purpose is to make visible the treatments that are produced in the system.

• News of the Year:
  Implementation of new analyses, facilitation of automatic use on various corpora of spontaneous speech.
• URL:
  https://­academics.­slodim.­loria.­fr/
• Contact:
  Maxime Amblard
• Partners:
  Loria, Université de Lorraine, CNRS

7.1.4 HostoMytho

• Keywords:
  Semantic annotation, Annotation tool, Semantic, Medical applications, NLP
• Functional Description:
  HostoMytho is a GWAP, or "game with a purpose" developed within the framework of the CODEINE ANR project. The aim of the game is to allow users to annotate medical files generated automatically, in order to evaluate their plausibility (quality of the language and medical semantics) and to add different layers of information (negation, hypothesis, time, etc.). HostoMytho is multiplatform.
• Contact:
  Bertrand Remy
• Partners:
  LISN, CEA-List

7.1.5 Arborator-Grew

• Name:
  Arborator's Collaborative Annotation
• Keywords:
  Annotation tool, Syntactic analysis
• Functional Description:
  The online interface allows managing collaborative annotation projects in dependency syntax. It is possible to use Grew queries and also to directly rewrite graphs in the annotation tool.
• News of the Year:
  Different features have been developed during 2023: redefinition of different types of user roles and access, integration of a labeling system in order to organize the annotation workflow, implementation of a full synchronization between Github and ArboratorGrew, improvement of different existing features (such as Lexicon, Grew search and Grew rule edition), enhancement of the parser UI, fixes for bugs and issues reported by the users, development of user documentation.
• URL:
  https://­arborator.­github.­io/
• Publication:
  hal-03021720
• Contact:
  Bruno Guillaume
• Participants:
  Khensa Amani Daoudi, Bruno Guillaume, Gael Guibon, Kim Gerdes, Kirian Guiller
• Partners:
  Université Paris Nanterre, LIMSI, LISN

7.2.1 Morphosyntactic Treebanks

Participants: Bruno Guillaume, Valentin Richard, Guy Perrier.

Sémagramme actively participates in the maintenance and the development of morphosyntactic treebanks in the Surface Syntactic Universal Dependencies (SUD) and Universal Dependencies (UD) projects.

Several French treebanks are involved:

• SUD_French-GSD
• SUD_French-ParisStories
• SUD_French-Rhapsodie
• SUD_French-Sequoia

A treebank of French sentences containing interrogative clauses, called French Interrogative Bank (UD_French-FIB) was extracted from French UD treebanks using the program FUDIA, designed for this purpose.

In 2023, by collaborating with linguists and field linguists in the Autogramm project, Sémagramme is also helping to develop treebanks for new languages (Haitian Creole and Hausa) and to improve and extend existing treebanks (Beja, Naija and Zaar).

• In Haitian Creole: SUD_Haitian_Creole-Autogramm
• In Hausa: SUD_Hausa-Autogramm
• In Beja: SUD_Beja-NSC
• In Naija: SUD_Naija-NSC
• In Zaar: SUD_Zaar-Autogramm

8.1.1 Abstract Categorial Grammars

8.1.2 Formal semantics of dependency relations

We have laid the foundations of a formal semantic theory for dependency grammars 42 by exploiting the following principles:

1. Dependency relations are represented as binary functions of type $\alpha \to \beta \to \alpha$ (or $\beta \to \alpha \to \alpha$), where $\alpha$ is the syntactic category of the governor and $\beta$ is the syntactic category of the governee.
2. By virtue of a coherence principle, the different ways of encoding a dependency structure by means of a $\lambda$-term should all give rise to the same semantic interpretation.
3. The semantic interpretation of a syntactic category cat is a type of the form ${\beta }_1\to \cdots {\beta }_n\to \alpha$, where $\alpha$ is the Montagovian interpretation of cat and ${\beta }_1,...{\beta }_n$ are the Montagovian interpretations of the syntactic categories of the phrases whose heads can potentially be governed by the head of a phrase of category cat.
4. Saturating operators allow phrases to recover their usual Montagovian interpretations.
5. Verbs and verbal phrases are semantically interpreted as sets of sets of events.

The resulting theory is fully compositional, allows for a treatment of quantification, and is robust in the sense that it supports the interpretation of partial dependency structures. We have then demonstrated that our theory allows one to deal with more advanced syntactic phenomena 38. We considered two cases: the relative clauses (which depend on the acl:relcl dependency relation) and the open clausal complements (which depend on the xcomp dependency relation).

8.1.3 Lexical Semantics and Linguistic Knowledge

Discourse Markers are lexical items which, like so or well in English, are used to organize discourse or to manifest epistemic or affective states of the speaker. Their semantic contribution is difficult to characterize in a sufficiently general way, in particular because they denote entities with a rich semantic structure, such as beliefs or speech acts. Preliminary results suggest that it is necessary to view their semantic update potential as involving intentions, actions and, generally speaking, dynamic modalities. The question then arises of what formal languages/systems are the most appropriate to capture these features (modal logic? Belief-Desire-Intention systems, interactive semantics à la Ginzburg, monads to mimic side effects, etc.).

8.1.4 Semantic Representation

We have been studying and comparing different existing semantic representation formalisms (Abstract Meaning Representation (AMR), Discourse Representation Theory (DRT), Uniform Meaning Representation (UMR), to name but a few). The aim is to determine, through theoretical and empirical studies, whether these formalisms are compatible and if they encode the same level of semantic information. This gives us a better understanding to open a discussion on what we want semantic representation to be, and build the groundwork for a unifying formalism.

In 24, an empirical comparison of two of the most popular semantic representation formalisms, AMR and DRS, is presented. Its goal is to explore the viability of transforming graphs from one framework into another to construct parallel datasets, but also to obtain an understanding of the similarities and differences through the use of annotated data. As no freely available parallel datasets that are common between the two exist, a sample of 200 sentences already annotated in DRS was taken and also annotated in AMR. A corpus-based approach was then used to build a graph rewriting system from DRS to AMR. This gives insights into where the two formalisms differ in encoding various semantic phenomena.

In 25, a set of structural (encoding of negation, temporal information, modality, etc.) and global (universality, scalability, etc.) features to consider when designing formalisms, and against which formalisms can be assessed, is proposed. A comparison of eight semantic representation formalisms, some that are graph-based and some that are logic-based, across the proposed features is provided, complemented by a comparison across the features of a more entailment-oriented corpus (FraCaS).

In 32, a first version of a layered approach to semantic representation is presented. It is motivated by the need for a semantic representation formalism which can encode a rich variety of semantic phenomena, while remaining simple to annotate and easy to read. The representation derives its core predicate-argument structure from Abstract Meaning Representation (AMR). The structure is then extended to encode "features", where each feature is considered as a layer representing a semantic phenomenon that is to be encoded. This approach allows for the possibility to exclude layers from the representation easily, while still being able to represent phenomena such as scope that are difficult for most graph-oriented formalisms.

8.1.5 Semantics of questions

Natural language statements are composed not only of declarative sentences but also of interrogative ones. Moreover, sentences cannot be categorized into purely declarative or purely interrogative sentences. Typically, a declarative statement may contain an indirect interrogative clause:

• (a)
  I don't know where Mary is.

Conversely, a direct interrogative clause may contain a declarative subordinate:

• (b)
  Do you know that Mary is here?

This interaction between declarative and interrogative clauses is particularly present in dialogues, where the logical notion of answerhood is as significant as the one of inference.

In order to tackle this issue from a formal standpoint, we investigated the properties and possible uses of inquisitive semantics, which is a formal semantic theory based on a logic that provides a uniform treatment of both declarative and interrogative expressions. Valentin Richard is currently working on a semantic model for the presuppositions and dynamic referential effects of interrogative pronouns. Some further investigation is required to understand the interaction between these presuppositions and possibility modals.

8.2.1 Dialogue Modeling

Together with Chloé Braud (IRIT), Maxime Amblard and Chuyuan Li continued to work on modeling discourse. Discourse processing suffers from data sparsity, especially for dialogues. As a result, they have explored approaches to build discourse structures for dialogues, based on attention matrices from pretrained language models (PLMs). They have investigated multiple tasks for fine-tuning and show that the dialogue-tailored Sentence Ordering task performs best. To locate and exploit discourse information in PLMs, they have proposed an unsupervised and a semi-supervised method. The proposals thereby achieve encouraging results on the STAC corpus, with F1 scores of 57.2 and 59.3 for the unsupervised and semisupervised methods, respectively. When restricted to projective trees, the scores improved to 63.3 and 68.1, see 23.

They also start working on the design of a full discourse parser 22. Discourse analysis plays a crucial role in Natural Language Processing (NLP) and has demonstrated its usefulness in various downstream applications like summarization and question answering. In this work, they studied discourse in dialogues: an under-explored setting due to significant data scarcity challenge. They conducted discourse parsing within a pipeline: first predicting the discourse structure, and then identifying the relations within the structure. Using only 50 examples as gold training data, the methods achieve competitive results compared to supervised state-of-the-art in-domain and much stronger performance cross-domain, with also better stability.

Maxime Amblard and Amandine Decker continue to work on topic shift modeling 17. Topics play an important role in dialogue coherence, as what is currently discussed constrains the possible contributions of the participants, and initiating a topic while the previous one is still under discussion may be confusing without appropriate signals. However, the way to actually define the notion of topic is debated in linguistics and not sufficiently discussed in dialogue modeling. A precise description of topics and topic shifts in conversation would contribute to a better understanding of what makes us judge a sequence of utterances to be coherent. In order to analyze different types of topic shifts, they proposed to create a corpus of written task-oriented conversations (discussion of the ethical dilemma of the balloon task), where the dialogues happen by message exchanges. Such a controlled setting where the main topic is fixed, and subtopics are more easily identifiable, could be very helpful when it comes to understanding how people change the topic and react to topic shifts in dialogues.

8.2.2 Discourse Markers

Jacques Jayez is currently working with Mathilde Dargnat (ATILF), Paola Herreño (Ph. D. candidate ATILF-LLF) and Maeva Sillaire (Ph. D. candidate ATILF) on the semantic representation of D(iscourse) M(arkers). DM are words/expressions like so or well in English which help structuring discourse or communicating speakers' internal epistemic or affective states. The domain-based approach initiated in the 90s consists in defining different types (aka domains) of semantic objects, like states of affairs, beliefs or speech acts. Such objects have a rich internal structure, which calls for a sufficiently expressive representation. For instance, speech acts involve discourse planning. Moreover DM can impact various layers of meaning (propositional content, presupposed content, etc.). Our goal is to implement formally the domain intuition by using tools from languages with a flexible subtyping mechanism (Ocaml) and to investigate whether the notion of monad, familiar in Haskell, can help us to characterize some `side effects' of DM.

Jacques Jayez is also working on the argumentative dimension of discourse, using a combination of standard Bayesian approach and game-theoretical notions 48, 39.

8.2.3 Pathological Discourse Modeling

Michel Musiol is once again part-time delegate at Inria (délégation nationale SHS) for the period 2023-2024. This proximity has enabled us to pursue an active collaboration on the modeling of pathological and clinical discourse.

In the context of the MePheSTO project (Digital Phenotyping for Psychiatric Disorders from Social Interaction - DFKI-Inria AI project), a multimodal perspective on this issue has been introduced. This includes the development of an interlocutory model of what might be termed a "therapeutic effect" in clinical/psychopathological interviewing, with regard to the cognitive-discursive profile and oculomotor behavior of participants (Amandine Lecomte ś thesis forthcoming 2024).

Indeed, based on a previous study addressing the issue of cognitive impairment bypass based on the dynamics of the repetition process  59 and another modeling visual attention in the psychologist-schizophrenic patient interaction  66 as well as on the basis of new data (for example, from subjects with ultra-high risk of developing psychosis or subjects who have experienced a major depressive episode 12, in the MePheSTO project), Maxime Amblard, Michel Musiol and Amandine Lecomte have explored the dependent relationships between supportive visual behavior and interaction dynamics, on the one hand, and the relationships between mental pathology and oculomotor disorders, on the other hand.

In addition, in  64 they have sketched a methodology for analyzing speech disorders, which will have the particularity of helping to select the discontinuous sequences most likely to carry thought disorders. They have anticipated the development of a modeling system based on the principles of pragmatic linguistics and formal semantics, which, when applied to carefully selected discontinuous discourse sequences  63, will have a good chance of revealing the nature of the underlying thought disorders. They compared the conjectures with the results of an earlier study on the discovery of four "proven" types of discontinuous sequences, and showed which of these sequences can therefore be considered carrying thought disorders.

They also analyzed these sequences by testing certain principles of semantic modeling in order to identify the nature of the disorders and thought operations underlying the discontinuous sequences concerned  65, 64. They show that discursive thought disorders should not be considered simply as the expression of a dysexecutive syndrome, but also as a device likely to affect more complex thought operations such as inferences involved in the conversational context representation system, in semantic memory and in calculating the meaning of utterances or in calculating the meaning of the speaker.

Improving the heuristics of formal systems for recognizing speech disorders and interpreting thought disorders on the basis of more appropriate and accurate semantic modeling may lead to the development of more discriminating and effective diagnostic tools. The formatting of the formal systems they have achieved will make it possible to represent the interlocutory structure of the disorder more and more accurately in its natural context of expression (speech), and should lead to the development of computerized diagnostic aids.

Finally, the increased precision of formal modeling applied to communication disorders should also make it possible to test the hypothesis that certain discourse configurations are related to thought disorders (in the broad sense), while others reveal cognitive dysfunctions that have more to do with the conditions of the possibility of discourse.

In this line, Michel Musiol has also updated MOS-SF36 norms in the young French population  68.

This program focuses on accurate and rapid diagnosis, as well as long-term therapeutic follow-up. These are major challenges for contemporary psychopathology. Thanks to modeling  59, 60 and computing  54, this work is developing multimodal methodologies for investigating symptoms (language, speech, neuropsychological and cognitive processes, eye movements and visual attention) that are sufficiently accurate to give rise in the medium term to the development of computerized diagnostic and therapeutic follow-up tools for the benefit of those involved in mental health.

8.2.4 Cognitive traces of side issues

It is by now widely believed that natural language communication operates at several levels. This means that information is distributed across several partially independent dimensions. For instance, a sentence like My stupid neighbor made noise again, simultaneously conveys that my neighbor made noise (the truth-conditional content), that the speaker considers he is stupid (an expressive, side issue 1) and that he had made noise before (the presupposition of again, side issue 2). While these phenomena have been extensively described from an empirical perspective, there is at present no unified framework for representing their differences and possible interactions under a formal, computational or cognitive point of view.

We examined the motor effects of presuppositions, using the convenient lexical material of factive verbs, that is, verbs which presuppose the truth of the complement clause. For example, Mary knows that Paul cheated on the exam presupposes that Paul cheated on the exam (side issue) and asserts (truth-conditional content) that she believes that. It has been shown that the oral presentation of movement-related verbs like jump or push elicits some activation in the motor cortex and finally results into an involuntary contraction of the thumb-index arc, which can be recorded by a special electromagnetic cell, called a grip force sensor.

We adapted this technique to the case of factive verbs on a series of sentences of the form Mary knows that Paul throws the ball, compared with high base-level sentences like Paul throws the ball and low base-level sentences like Paul does not throw the ball. Summarizing, our results indicate that the sentences with the factive verbs elicit a very similar response to that of high base-level ones, and, as expected, a different response from that of low base-level one. This suggests that, at least for factive verbs, the presupposed status leaves no trace of a special cognitive treatment, which would lead for instance, to a delayed or weaker motor response.

However, when applied to more complex negative sentences like Mary does not know that Paul throws the ball, there is no evidence of a motor trace. This is in agreement with observations in the literature suggesting that, under negation, presuppositions are more 'difficult' to process than in simple assertive sentences. More precisely, in the case of motor response, negation interacts with the presupposition, which suggests that truth-conditional content and side issue cognitive processing cannot be totally separated 8.

8.3.1 Universal Dependencies and Surface Syntactic Universal Dependencies

The Universal Dependencies project (UD) aims at building a syntactic dependency scheme which allows for similar analyses for several different languages. Bruno Guillaume and Guy Perrier are active in the UD community, and participate in the development and the improvement of the French data in this international initiative. In 2023, two new versions (2.12 and 2.13) of the UD data were released.

During 2023, they continued working, in collaboration with Sylvain Kahane, Kim Gerdes and their teams on the promotion of the Surface Syntactic Universal Dependencies (SUD) framework. SUD is an annotation scheme for syntactic dependency treebanks, that is almost isomorphic to UD (Universal Dependencies). Contrary to UD, it is based on syntactic criteria (favoring functional heads) and the relations are defined on distributional and functional bases. In 41, they describe how function words are considered in the SUD formalism and compare with other approaches.

A suggestion to add clause type features and change the annotations of some French expressions regarding interrogatives was investigated by Valentin Richard 33.

8.3.2 Multiword annotation in the Parseme project

In 40, the new version 1.3 of the PARSEME multilingual corpus annotated with verbal multiword expressions is presented. Since the previous version, new languages have joined the undertaking of creating such a resource, some of the already existing corpora have been enriched with new annotated texts, while others have been enhanced in various ways. The PARSEME multilingual corpus represents 26 languages now. All monolingual corpora therein use Universal Dependencies v.2 tagset. They are (re-)split observing the PARSEME v.1.2 standard, which puts an impact on unseen VMWEs. With the current iteration, the corpus release process has been detached from shared tasks; instead, a process for continuous improvement and systematic releases has been introduced.

During the MWE workshop, we showed 20 how Grew and Grew-match can be adapted to MWE annotated data of the Parseme project. The tool can be used for linguistic exploration on the data, for helping the manual annotation process and to search for errors or inconsistencies in the annotations.

8.3.3 Induction of Descriptive Grammars

The ANR project Autogramm (Induction of descriptive grammar from annotated corpora) started in 01 2022. The goal of this project is to automate, as far as possible, the extraction of descriptive grammars and grammatical descriptions from annotated corpora for linguistic and typological studies. The project also promotes the development of treebanks for low-resourced languages, in order to extract quantitative descriptive grammars for these languages.

In 31, the project was presented to the French community. Treebanks for new languages were also released during 2023, one in Zaar and one in Haitian Creole.

8.3.4 Mapping Lexical Resources

Lexical resources are essential for the development of tools and methods for the various tasks of NLP. In French, the existing resources are heterogeneous in their size, their construction and their level of linguistic description, which opens the way to explore a method to group or link/map them automatically. French lexical resources are plentiful, but have not all been used in the same way. In 27, Hee-Soo Choi she presented a diachronic analysis of 34 French lexical resources over a 20-year period in the proceedings of the TALN conference, showing differences in the types of application and reuse of resources by the community.

8.3.5 Sentence Semantic Similarity Corpus

In the context of the CODEINE ANR project and more specifically of Nicolas Hiebel's PhD thesis, Karën Fort worked with Aurélie Névéol (LISN-CNRS) and Olivier Ferret (CEA) on the creation of a sentence semantic similarity corpus for French on the clinical domain. A paper on the subject was presented at EACL 2023 21, a rank A conference of the domain. An adaptation in French of the paper was accepted at TALN 2023 29.

The key idea of the project is to use confidential corpora to automatically generate anonymous synthetic texts capable of emulating real documents from the perspective of their linguistic characteristics. The project will rely on a Games With A Purpose and crowd sourcing to validate and then annotate these synthesized clinical texts.

This game, developed by Bertrand Remy, is called HostoMytho (see Section 7.1.4), and includes various mini-games for different annotation layers, such as negation, error typing, or plausibility rating. The game is multi-platform, and therefore intended to be used on the web, on Android and iOS.

8.4.1 Ethics@Loria

Karën Fort originated a working group at LORIA for AI ethics (ethics@loria), involving researchers from various teams, including Maxime Amblard, Marc Anderson, Armelle Brun (BIRD), Mathieu d'Aquin (Orpailleur), Christophe Cerisara (Synalp), Anne Bonneau (Multispeech), Slim Ouni (Multispeech) and Abdessamad Imine (Pesto). Aurore Coince helped manage the group. Ethics@loria proposed the Doctoral training on Ethics "Write your dystopia".

8.4.2 Evaluating Stereotypes in Masked Language Models in Many Languages

Following the creation of French CrowsPairs 6, Karën Fort contacted researchers interested in creating a CrowsPairs corpus for their language, in order to test the language models. The group got bigger and bigger to finally include 22 researchers (including Fanny Ducel, from the team) for 7 languages (German, Maltese, Spanish, Italian, Chinese, standard Arabic and Catalan). The group worked together for more than a year to produce corpora and test masked language models in these 7 new languages. The corpora are freely available, along with the code to test the language models and the guidelines we followed for the adaptation.1 It's important to notice that this work has been performed without any funding. The work has been detailed in a paper submitted at LREC-COLING 2024.

8.4.3 Evaluating stereotypes in autoregressive language models

Fanny Ducel authored a critical literature review on the topic of stereotypical biases in language models, under the supervision of Karën Fort and Aurélie Névéol. She presented this work at the Workshop on Algorithmic Injustice in Amsterdam in June 2023 36. This study was part of her internship and master's thesis on the evaluation of stereotypical biases in autoregressive language models, that has become her PhD topic since October 2023.

Karën Fort is PI of a new 4 year ANR project (2023-2027), InExtenso (Intrinsic and Extrinsic evaluation of biases in large language models), in collaboration with Rouen's hospital (CHU) and LISN-CNRS. The project aims at better identifying stereotyped biases in LLMs in French and, when possible, mitigate them. At LORIA, the project reaches beyond Sémagramme, as Miguel Couceiro (Orpailleur) joined us on the bias mitigation part.

8.4.4 NLP for NLP and Ethics

Karën Fort originated a research group on the impact of the BigTech companies on NLP (industry presence, potential thematic shifts, participation in paper authorship). This group is composed of Aurélie Névéol (LISN), Saif Mohammad (National Research Council Canada), Mohamed Abdalla (University of Toronto), Terry Lima Ruas and Jan Philip Wahle (Wuppertal University) and Fanny Ducel (M2 student at Sorbonne University). The group regularly gathered from 09 to 12 2022 to perform both an automatic study on all the ACL Anthology papers and a manual one from the ACL 2022 papers. The resulting paper was accepted at ACL 2023 15, the A* conference in NLP.

8.4.5 Ethics in AI Integration into Industry

Karën Fort and Marc Anderson completed the Ethics by Design component of the EU Horizons AI-Proficient project, which ended in October 2023. The final year of the project involved a review of ethics integration in the project in comparison to the EU High Level Expert Group Guidelines, as well as an analysis of the results of the implementation of ethical recommendations given to project partners. That analysis was aimed at providing a model for future projects and AI ethics researchers for operationalizing AI ethics, which remains very rare in the domain. Contributions were made to various late-stage deliverables of the project, namely those of the WP6 which dealt with project validation, and in particular the ethics team provided the AI-Proficient Deliverable 6.4: AI-Proficient Ethical Recommendations.

Marc Anderson 11 authored a journal paper in Studia Philosophica Wratislaviensia carrying out an analysis of public ethical concerns and expectations for AI as compared to the older technology of the telegraph. The latter paper was initially presented as “AI Ethics and the Lessons of History,” in the 2nd International Conference on the Ethics of Artificial Intelligence (2ICEAI)34. Marc Anderson also presented at the Sohoma 2023 conference 43, on the topic of ethical sustainability in digital manufacturing and industry 4.0 generally.

The ongoing collaboration between Karën Fort and Marc Anderson resulted in no new common article in 2023 (one has been accepted and will be published in 2024, though), but Karën Fort has been invited as a keynote at ERGO IA 13 to talk about how the recommendations made by the ethics group were taken into account in the AI-Proficient project.

9.1.1 Yseop

Participants: Philippe de Groote, Maxime Guillaume, Sylvain Pogodalla.

The Sémagramme team has set up a Cifre thesis contract with Yseop on ACG extensions and use in an industrial environment.

10.1.1 Visits to international teams

10.2.1 H2020 projects

10.2.2 Other european programs/initiatives

• Bruno Guillaume is a member of the core group of the cost action: CA21167 - Universality, diversity and idiosyncrasy in language technology (UniDive). He is the leader of the working group named "Corpus Annotation".

10.3.1 ANR Project: InExtenso

Participants: Karën Fort, Maxime Amblard, Michel Musiol, Fanny Ducel.

• Title:
  Intrinsic and Extrinsic evaluation of biases in large language models
• Duration:
  10 2023–09 2027
• Coordinator:
  Karën Fort
• Partners:
  CHU Rouen, LISN, LORIA
• Participants:
  Maxime Amblard, Fanny Ducel, Karën Fort (coordinator), Michel Musiol, Miguel Couceiro
• Abstract:
  Large Language Models (LLM) are the Swiss Army knife of today’s Natural Language Processing (NLP). They often outperform the state-of-the-art on benchmarks commonly used in the field for tasks such as part-of-speech tagging, text classification and named-entity recognition, thus paving the way to a myriad of end-user applications. However, it has been shown that LLM exhibit major ethical issues including significant environmental impact, mirroring and amplification of stereotyped biases, which in turn have a disproportionate impact on historically disadvantaged social groups. It is urgent to address the social impact of NLP as the applications we develop, such as chatGPT, are now directly made available to end users. The detection and mitigation of biases have therefore become an active area of research in the past few years, focusing mainly on Masked Language Models (MLM) such as BERT in English and the North American social context. Several sources of bias were identified in the NLP pipeline. However the interconnection between sources and overall impact of each source on downstream applications remains unclear. In this project, we want to observe the entire pipeline, from the intrinsic point of view (within the model itself), to the pre-training task point of view (in the case of autoregressive LLM, text generation), on to some real-world downstream applications. We chose to focus on two types of medical applications: mental illness diagnosis help and information extraction from clinical records for public health purposes such as patient enrollment into clinical trials. The project will provide corpora and methods for a global evaluation of bias in LLM in French as well as studies to further the understanding of biases in clinical NLP pipelines and the environmental impact of the integration of these models in digital health.

10.3.2 ANR Project: CoDeinE

Participants: Karën Fort, Bruno Guillaume, Bertrand Remy.

• Title:
  artificial text COrpus DEsIgNed Ethically automatic synthesis of clinical documents
• Duration:
  03 2021–02 2025
• Coordinator:
  Aurélie Névéol (Limsi)
• Partners:
  CRC, CEA List, LISN, LORIA
• Participants:
  Bruno Guillaume, Karën Fort (local coordinator), Bertrand Remy
• Abstract:
  Machine learning methods have become prevalent in language technologies. They rely on annotated corpora to train models and evaluate algorithms. The CoDeinE project proposes to address the lack of shareable corpora in sensitive domains such as health or banking. The key idea of the project is to use confidential corpora to automatically generate synthetic texts that mimic the linguistic properties of real documents while preserving confidentiality. We will use clinical documents in electronic patient records as a case study. Furthermore, the project will rely on Games With A Purpose and crowd sourcing to validate and annotate the synthesized texts.

10.3.3 ANR Project: Autogramm

Participants: Bruno Guillaume, Karën Fort, Guy Perrier, Khensa Amani Daoudi.

• Title:
  Induction of descriptive grammar from annotated corpora
• Duration:
  01 2022–12 2025
• Coordinator:
  Sylvain Kahane (Université Paris Nanterre)
• Partners:
  MoDyCo, LACITO, LISN, Inria Nancy – Grand Est
• Participants:
  Bruno Guillaume (local coordinator), Karën Fort, Guy Perrier
• Abstract:
  The goal of this project is to automate, as far as possible, the extraction of descriptive grammars and grammatical descriptions from annotated corpora for linguistic and typological studies. The project also promotes the development of treebanks for under-endowed languages, in order to extract quantitative descriptive grammars for these languages. The project uses the annotation scheme SUD (Surface-syntactic Universal Dependencies), the query tool Grew-match and the annotation tool ArboratorGrew.

10.3.4 ANR Project: CODIM

Participants: Maxime Amblard, Jacques Jayez.

• Title:
  Compositionality and discourse markers
• Duration:
  01 2023–12 2026
• Coordinator:
  Mathilde Dargnat (Université de Lorraine and ATILF)
• Partners:
  ATIL, LLF, Inria Nancy – Grand Est
• Participants:
  Maxime Amblard, Jacques Jayez
• Abstract:
  The CODIM project focuses on the two main linguistic resources for organizing monologues or conversations in human languages : D(iscourse) M(arkers)(therefore/donc, well/ben, bon etc. in English/French) and prosody (in particular, intonation). It will evaluate their status with respect to two major views on communication: compositionality (the possibility of combining meaningful expressions into more complex meaningful expressions) and pattern or construction-based approaches (the idea that language users exploit partly `frozen’ strings of words). We will compare the semantic and prosodic properties of simple and complex French DM (e.g. ah + bon) found in corpora for written and spoken French, using a variety of technical tools for DM identification (category-driven text mining), clustering (statistics and Machine Learning) and research in prosody (duration and intensity measures, contour representation). The project fosters a number of collaborations between linguists and computer scientists.

11.1.1 Scientific events: organisation

• Maxime Amblard: main organiser of the 15th International Conference on Computational Semantics, (IWCS 2023), Nancy 20-23 06 2023.
• Valentin Richard: local chair of the workshop Inquisitiveness Below and Beyond the Sentence Boundary (InqBnB4), Nancy, 20 06 2023.
• The Sémagramme team organized a scientific memorial day in honor of their colleague Guy Perrier who passed away in 2023.

11.1.2 Scientific events: selection

11.1.3 Journal

11.1.4 Invited talks

• Karën Fort: Ethics by design for real: lessons learned from an industry 4.0 European project. ERGO IA, Biarritz, France, 12 10 2023 13.
• Guy Perrier: Why is graph rewriting interesting for computational linguistics?. Keynote presentation at GURT/SyntaxFest 2023, Washington D.C., 9 3 2023 14.

11.1.5 Invited seminars

• Hee-Soo Choi: Corpus-based Language Universals Analysis using Universal Dependencies. Team seminar Traitement Automatique du Langage Écrit et Parlé (TALEP), Laboratoire d'Informatique et Systèmes (LIS), Marseille, France, 23 03 2023.
• Bruno Guillaume: Introduction to Grew-match. 1st UniDive webinar, online, 19 6 2023.
• Valentin Richard: Extraction des Interrogatives de Corpus Francophones Annotés en Dépendances Universelles, Café TAL, Atilf, Nancy, 26 06 2023.

11.1.6 Leadership within the scientific community

• Maxime Amblard: Leader of OLKI2.0 project (Lorraine Université d'Excellence project - PIA).
• Karën Fort has been co-chair of the ACL Ethics Committee since 2021, with Min-Yen Kan (Univ. of Singapore) and Y. Tsvetkov (Univ. of Washington).
• Bruno Guillaume is Working Group leader in a CA21167 - Universality, diversity and idiosyncrasy in language technology (UniDive).

11.1.7 Scientific expertise

• Maxime Amblard: evaluation for the ANR generic call 2023.

11.1.8 Research administration

• Maxime Amblard:
  • Head of the master in Natural Language Processing (master 1 and 2).
  • Member of CNU 27 (Computer Science).
• Karën Fort:
  • Member of CNU 27 (Computer Science): participation to qualifications, promotions and suivi de carrière.
• Bruno Guillaume:
  • Head of the Natural Language Processing and Knowledge Discovery department of the LORIA laboratory.
  • Manager (with Alain Polguère) of the CPER (Contrat de Plan État-Région) Langues, Connaissances et Humanités Numériques.
• Sylvain Pogodalla:
  • Elected member of the comité de centre Inria Nancy – Grand Est.
  • In charge of the local commission IES (information et édition scientifique) of the Inria Nancy – Grand Est and LORIA.
  • Member of the national commission IES of Inria.

11.2.1 Teaching

• Licence:
  • Maxime Amblard, AI Introduction, 36h, L1, Université de Lorraine, France.
  • Maxime Amblard, Chuyuan Li and Marie Cousin NLP for beginners, 20h, L2, Université de Lorraine, France.
  • Maxime Amblard and Chuyuan Li, Linguistic engineering, 20h, L3, Université de Lorraine, France.
  • Hee-Soo Choi, Databases, 14h, Telecom Nancy, Université de Lorraine, France.
  • Amandine Decker, Research Methods in Linguistics, 10h, University of Gothenburg, Sweden.
  • Karën Fort, Relational databases, 55h, L3, Sorbonne Université, France.
  • Pierre Ludmann, Informatics 2, 27h, Mines Nancy, Université de Lorraine, France.
  • Pierre Ludmann, Databases, 114h, Polytech Nancy, Université de Lorraine, France.
  • Pierre Ludmann, Java project, 16h, Polytech Nancy, Université de Lorraine, France.
  • Pierre Ludmann, Year-long project, 20h, Polytech Nancy, Université de Lorraine, France.
  • Pierre Ludmann, Mentoring, 40h, Polytech Nancy, Université de Lorraine, France.
  • Siyana Pavlova, Algorithmique - Programmation 1, 44h, L1, Université de Lorraine, France.
  • Siyana Pavlova, Bases de données, 28h, L2, Université de Lorraine, France.
  • Valentin Richard, Langages de Scripts, 50h, L3, Université de Lorraine, France.
  • Vincent Tourneur, Structures de données, 40h, L1, IUT Charlemagne, Université de Lorraine, France.
• Master:
  • Maxime Amblard, Siyana Pavlova and Fanny Ducel, Python Programming, 30h, M1 NLP (IDMC), Université de Lorraine, France.
  • Maxime Amblard and Siyana Pavlova, Methods for NLP, 36h, M1 NLP (IDMC), Université de Lorraine, France.
  • Maxime Amblard, NLP project, 20h, M1 NLP (IDMC), Université de Lorraine, France.
  • Maxime Amblard, Marie Cousin, and Amandine Decker, Formalisms and Syntax, 24h, M2 NLP (IDMC), Université de Lorraine, France.
  • Maxime Amblard, Valentin Richard and Siyana Pavlova, Discourse and Dialogue, 18h, M2 NLP (IDMC), Université de Lorraine, France.
  • Marie Cousin, Agile Method and Scrum, 4h, M2 NLP (IDMC), Université de Lorraine, France.
  • Marie Cousin and Amandine Decker, Data Structures, 20h, M1 NLP (IDMC), Université de Lorraine, France
  • Philippe de Groote, Formal Logic, 22h, M1 NLP (IDMC), Université de Lorraine, France.
  • Philippe de Groote, Formal languages, 22h, M1 NLP (IDMC), Université de Lorraine, France.
  • Philippe de Groote, Computational Semantics, 18h, M2 NLP (IDMC), Université de Lorraine, France.
  • Philippe de Groote, Computational structures and logics for natural language modeling, 18h, M2 NLP (IDMC), Université Paris Diderot – Paris 7, France.
  • Karën Fort, Data Ethics, 18h, M1 ISF (Informatique et Statistique financières), Université Panthéon Assas, France.
  • Karën Fort, Ethics and NLP (English), 17h30, M2 NLP (IDMC), Université de Lorraine, France.
  • Karën Fort, Formal Grammar, 39h, M1, Sorbonne Université, France.
  • Karën Fort, Corpora, resources and tools for linguistics, 39h, M1, Sorbonne Université, France.
  • Karën Fort, Ethics and NLP, 15h, M2, Sorbonne Université, France.
  • Karën Fort, Collaborative annotation for NLP, 30h, M2, Sorbonne Université, France.
  • Bruno Guillaume, Written Corpora TAL (English), 45h, M1 NLP (IDMC), Université de Lorraine, France.
  • Chuyuan Li, Introduction to G5K (English), 4h, M1 NLP (IDMC), Université de Lorraine, France.
  • Pierre Ludmann, Software Engineering 2, 10h, Mines Nancy, Université de Lorraine, France.
  • Pierre Ludmann, Compilation, 27.5h, Mines Nancy, Université de Lorraine, France.
  • Pierre Ludmann, Department project, 5h, Mines Nancy, Université de Lorraine, France.
  • Pierre Ludmann, Software Engineering, 42h, Polytech Nancy, Université de Lorraine, France.
  • Pierre Ludmann, Web project, 6h, Polytech Nancy, Université de Lorraine, France.
  • Pierre Ludmann, End-of-study internships, 20h Polytech Nancy, Université de Lorraine, France.
  • Pierre Ludmann, End-of-study project, 20h Polytech Nancy, Université de Lorraine, France.
  • Pierre Ludmann, Mentoring, 48h, Polytech Nancy, Université de Lorraine, France.
  • Vincent Tourneur, UML beginners (English), 10h, M1 NLP (IDMC), Université de Lorraine, France.
• Doctorate:
  • Karën Fort, Scientific integrity, 6h, École doctorale 5, Faculté des lettres, Sorbonne Université.
  • Karën Fort, Ethics and biases in NLP, Summer school of the GDR TAL, ETAL 06 2023 – Marseilles, France.
• International Summer School:
  • Bruno Guillaume with Kim Gerdes, Treebanking: methodology, tools and applications at the 34th European Summer School in Logic, Language and Information (ESSLLI 2023) in Ljubljana, Slovenia.
• International Tutorials:
  • Karën Fort with Lucianna Benotti (Universidad Nacional de Córdoba, Argentine), Yulia Tsvetkov and Min-Yen Kan. Tutorial at EACL 2023 (CORE A): Understanding Ethics in NLP Authoring and Reviewing (Introductory) 51.

11.2.2 Supervision

• PhD
• Chuyuan Li, Formal and Statistical Modeling of Dialogue, since 10 2019. Supervision: Maxime Amblard and Chloé Braud (IRIT).
• PhD in progress
  • Samuel Buchel, Linguistic, Semantic and Cognitive Modeling of Dialogical Incongruities and Discontinuities in The Interaction with The Schizophrenic Patients, since 12 2019. Supervision: Maxime Amblard and Michel Musiol.
  • Hee-Soo Choi, Lier des ressources lexicales du français en vue d’une interopérabilité entre niveaux linguistiques, since 10 2021. Supervision: Karën Fort and Mathieu Constant.
  • Marie Cousin, Modélisation de paraphrase dans les grammaires catégorielles abstraites, since 10 2022. Supervision: Philippe de Groote and Sylvain Pogodalla.
  • Amandine Decker, Modelling Topic-level Interaction in Pathological Conversations, since 10 2022. Supervision: Maxime Amblard and Ellen Breitholtz (University of Gothenburg, Sweden).
  • Fanny Ducel, Evaluating stereotyped biases in auto-regressive language models, since 10 2023. Supervision: Karën Fort and Aurélie Névéol (LISN-CNRS).
  • Maxime Guillaume, Structures de traits pour les Grammaires Catégorielles Abstraites, since 07 2021. Supervision: Philippe de Groote and Raphaël Salmon (Yseop).
  • Nicolas Hiebel, Création éthique de données textuelles artificielles : application au domaine biomédical, since 10 2021. Supervision: Aurélie Névéol (LISN-CNRS), Karën Fort and Olivier Ferret (CEA).
  • Amandine Lecomte, Analyse longitudinale de prise en charge psychothérapeutique de patients psychiatriques et de patients atteints de maladies neurodégénératives : informatisation et modélisation dialogique des indices comportementaux associés à l’efficacité (vs échec) des stratégies de prise en charge tentées par les thérapeutes, since 10 2019. Supervision: Michel Musiol and Alexandra König.
  • Pierre Ludmann, Dynamic Construction of Discursive Structures, since 09 2017. Supervision: Philippe de Groote and Sylvain Pogodalla.
  • Siyana Pavlova, Tools and Methods for Semantic Annotation, since 11 2020. Supervision: Maxime Amblard and Bruno Guillaume.
  • Valentin Richard, Aspects compositionnels et dynamiques de la sémantique inquisitrice, since 09 2021. Supervision: Philippe de Groote, Floris Roelofsen and Reinhart Muskens (Universiteit van Amsterdam, ILLC).
  • Priyansh Trivedi, Injecting Lexical and Semantic Knowledge into Word, Phrasal and Sentence Embeddings, since 11 2021. Supervision: Philippe de Groote and Pascal Denis (until 05 2023).

11.2.3 Juries

• Maxime Amblard. PhD reviewer (rapporteur) for Nesrine Bannour's PhD Information Extraction from Electronic Health Records: Studies on temporal ordering, privacy and environmental impact, Université Paris-Saclay, France, 3 11 2023.
• Maxime Amblard. PhD jury for Nicolas Devatine's PhD Discourse-Driven Prediction and Characterization of Textual Bias, Université Toulouse III - Paul Sabatier, France, 23 10 2023.
• Karën Fort. PhD reviewer (rapporteuse) for Nicolas Devatine's PhD Discourse-Driven Prediction and Characterization of Textual Bias, Université Toulouse III - Paul Sabatier, France, 23 10 2023.

11.3.1 Internal or external Inria responsibilities

• Maxime Amblard is a member of the scientific committee of )i( interstices.

11.3.2 Articles and contents

• Karën Fort. Comment les "BigTech" investissent la recherche en Traitement automatique des langues ?INS2I Web site, CNRS, France, 12 09 2023.
• Karën Fort. Interstices podcast. Quelle éthique pour les agents conversationnels ? 28 08 2023 53.

11.3.3 Education

• Marie Cousin and Amandine Decker: animation of a MATh.en.JEANS workshop within Edmond de Goncourt secondary school in Pulnoy.

11.3.4 Interventions

• Maxime Amblard. L’IA n’ira pas beaucoup plus loin. Oxford Style debate, Forum des Sciences Cognitves et du TAL, IDMC 29 11 2023.
• Maxime Amblard and Michel Musiol. Le repérage des signes de la pathologie mentale dans le discours, Cycle de conférences sur le Handicap, Inria, 6 04 2023.
• Maxime Amblard. Je parle à mon ordinateur, c'est grave docteur ?, cycle de conférence A votre santé, 16 03 2023.
• Karën Fort. Les enjeux éthiques de l’IA vus par le prisme du traitement automatique des langues. In Séminaire #3 IA: Enjeux éthiques et juridiques, ARCEP (Autorité de régulation des communications électroniques, des postes et de la distribution de la presse). Paris. 15 12 2023.
• Karën Fort, Hee-Soo Choi and Fanny Ducel: 2023-12-07, presentation to high school students in the context of the "Chiche !" initiative (lycée Varoquaux, Tomblaine, France).
• Karën Fort and Hee-Soo Choi: 2023-12-01, presentation to high school students in the context of the "Chiche !" initiative (lycée Claude Gellée, Épinal, France).
• Marie Cousin and Amandine Decker: 2023-02-01 and 2023-02-02, presentation to high school students in the context of the "Chiche !" initiative (lycée Cormontaigne, Metz, France).
• Marie Cousin, Amandine Decker and Vincent Tourneur: 2023-03-02 and 2023-03-03, presentation to high school students in the context of the "Chiche !" initiative (lycée Jean-Victor Poncelet, Saint-Avold, France).
• Amandine Decker: 2023-10-24, talks with high school students in the context of the first edition of the “Les cigognes” mathematics and computer science research retreat in the Grand-Est region.
• Valentin Richard gave a talk for Pint of Science at Taverne de l'Irlandais (Nancy) on 22 05 2023.
• Valentin Richard was interviewed by the newspaper L'Est Républicain about his participation to Pint of Science, 19 05 2023.
• Valentin Richard gave a talk 50 for the round-table discussion organized by CJD Section de Nancy at Télécom Nancy (Nancy), on 16 10 2023.

International journals

• 9 articleM. M.Marc M Anderson and K.Karën Fort. Evaluating the acceptability of ethical recommendations in industry 4.0: an ethics by design approach. AI & Society: Knowledge, Culture and CommunicationJanuary 2024HALDOI
• 10 articleM. M.Marc M Anderson. How we will Discover Sentience in AI.Journal of Social Computing432023HAL
• 11 articleM. M.Marc M Anderson. Rare Opportunity or History Revisited? The Pitfalls and Prospects for Ethical AI in light of Public Ethical Responses to the Telegraph.Studia Philosophica Wratislaviensia1832023HALback to text
• 12 articleE.Eric Ettore, P.Philipp Müller, J.Jonas Hinze, M.Matthias Riemenschneider, M.Michel Benoit, B.Bruno Giordana, D.Danilo Postin, R.Rene Hurlemann, A.Amandine Lecomte, M.Michel Musiol, H.Hali Lindsay, P.Philippe Robert and A.Alexandra König. Digital Phenotyping for Differential Diagnosis of Major Depressive Episode: Narrative Review.JMIR Mental Health10January 2023, e37225HALDOIback to text

Invited conferences

• 13 inproceedingsK.Karën Fort. Ethics by design for real : lessons learned from an industry 4.0 European project.ERGO'IABiarritz (France), FranceOctober 2023HALback to textback to text
• 14 inproceedings G.Guy Perrier. Why is graph rewriting interesting for computational linguistics? GURT/SyntaxFest 2023 Washington, United States March 2023 HAL back to text

International peer-reviewed conferences

• 15 inproceedingsM.Mohamed Abdalla, J. P.Jan Philip Wahle, T.Terry Ruas, A.Aurélie Névéol, F.Fanny Ducel, S. M.Saif M. Mohammad and K.Karën Fort. The Elephant in the Room: Analyzing the Presence of Big Tech in Natural Language Processing Research.Proceedings of the 61st Annual Meeting of the Association for Computational LinguisticsVolume 1: Long Papers61st Annual Meeting of the Association for Computational Linguistics1Toronto, CanadaAssociation for Computational Linguitics2023, 13141-13160HALback to text
• 16 inproceedingsM.Marie Cousin. Meaning-Text Theory within Abstract Categorial Grammars: Towards Paraphrase and Lexical Function Modeling for Text Generation.Proceedings of the 15th International Conference on Computational Semantics (IWCS)IWCS 2023 - 15th International Conference on Computational SemanticsNancy, FranceAssociation for Computational LinguisticsJune 2023HALback to text
• 17 inproceedingsA.Amandine Decker and M.Maxime Amblard. Analysing topic shifts in task-oriented dialogues.Journées scientifiques du GDR Lift - LIFT 2023Nancy, FranceNovember 2023HALback to text
• 18 inproceedingsA.Amandine Decker, E.Ellen Breitholtz, C.Christine Howes and S.Staffan Larsson. Topic and genre in dialogue.Proceedings of the 27th Workshop on the Semantics and Pragmatics of DialogueSEMDIAL 2023Maribor, SloveniaAugust 2023, 143-145HAL
• 19 inproceedingsI.Izaskun Fernandez, K.Kerman Lopez de Calle, E.Eider Garate, R.Regis Benzmuller, M.Melodie Kessler and M.Marc Anderson. Human-Feedback for AI in Industry.CENTRIC 2023The Sixteenth International Conference on Advances in Human oriented andPersonalized Mechanisms, Technologies, and ServicesCENTRIC 2023, The Sixteenth International Conference on Advances in Human-oriented and Personalized Mechanisms, Technologies, and ServicesCENTRIC 2023 The Sixteenth International Conference on Advances in Human oriented and Personalized Mechanisms, Technologies, and ServicesValencia, SpainNovember 2023HAL
• 20 inproceedingsB.Bruno Guillaume. Graph-based multi-layer querying in Parseme Corpora.Proceedings of the 19th Workshop on Multiword Expressions (MWE 2023)Dubrovnic, CroatiaAssociation for Computational LinguisticsJune 2023, 58-64HALDOIback to text
• 21 inproceedingsN.Nicolas Hiebel, O.Olivier Ferret, K.Karën Fort and A.Aurélie Névéol. Can Synthetic Text Help Clinical Named Entity Recognition? A Study of Electronic Health Records in French.The 17th Conference of the European Chapter of the Association for Computational Linguistics (EACL 2023)Dubrovnik, CroatiaMay 2023HALDOIback to text
• 22 inproceedingsC.Chuyuan Li, M.Maxime Amblard and C.Chloé Braud. A Semi-supervised Dialogue Discourse Parsing Pipeline.Journées Scientifiques du GDR Lift (LIFT 2023)Nancy, FranceNovember 2023HALback to text
• 23 inproceedingsC.Chuyuan Li, P.Patrick Huber, W.Wen Xiao, M.Maxime Amblard, C.Chloé Braud and G.Giuseppe Carenini. Discourse Structure Extraction from Pre-Trained and Fine-Tuned Language Models in Dialogues.Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics (EACL 2023)European Chapter of the Association for Computational Linguistics (EACL)Dubrovnik, CroatiaMay 2023, 2562–2579HALback to text
• 24 inproceedingsS.Siyana Pavlova, M.Maxime Amblard and B.Bruno Guillaume. Bridging Semantic Frameworks: mapping DRS onto AMR.Proceedings of The 15th International Conference on Computational Semantics (IWCS 2023)The 15th International Conference on Computational Semantics (IWCS 2023)Nancy, FranceJune 2023HALback to text
• 25 inproceedingsS.Siyana Pavlova, M.Maxime Amblard and B.Bruno Guillaume. Structural and Global Features for Comparing Semantic Representation Formalisms.Proceedings of the 4th International Workshop on Designing Meaning Representation (DMR 2023)The 4th International Workshop on Designing Meaning RepresentationNancy, FranceJune 2023HALback to text

National peer-reviewed Conferences

• 26 inproceedingsV.-T.Vincent-Thomas Barrouillet, M.Maxime Amblard and M.Michel Musiol. Toward an automatic identification of discontinuities in the pathological discourse of patient with schizophrenia.Journées scientifiques du GDR Lift - LIFT 2023Nancy, FranceNovember 2023HAL
• 27 inproceedingsH.-S.Hee-Soo Choi, K.Karën Fort, B.Bruno Guillaume and M.Mathieu Constant. Des ressources lexicales du français et de leur utilisation en TAL : étude des actes de TALN.Actes de CORIA-TALN 2023. Actes de la 30e Conférence sur le Traitement Automatique des Langues Naturelles (TALN), volume 2 : travaux de recherche originaux -- articles courts18e Conférence en Recherche d'Information et Applications -- 16e Rencontres Jeunes Chercheurs en RI -- 30e Conférence sur le Traitement Automatique des Langues Naturelles -- 25e Rencontre des Étudiants Chercheurs en Informatique pour le Traitement Automatique des LanguesParis, FranceATALA2023, 23-36HALback to text
• 28 inproceedingsM.Marie Cousin. Towards an implementation of meaning-text theory with abstract categorial grammars.Actes de CORIA-TALN 2023. Actes des 16e Rencontres Jeunes Chercheurs en RI (RJCRI) et 25e Rencontre des Étudiants Chercheurs en Informatique pour le Traitement Automatique des Langues (RÉCITAL)18e Conférence en Recherche d'Information et Applications -- 16e Rencontres Jeunes Chercheurs en RI -- 30e Conférence sur le Traitement Automatique des Langues Naturelles -- 25e Rencontre des Étudiants Chercheurs en Informatique pour le Traitement Automatique des LanguesParis, FranceATALA2023, 72-86HALback to text
• 29 inproceedingsN.Nicolas Hiebel, O.Olivier Ferret, K.Karën Fort and A.Aurélie Névéol. Les textes cliniques français générés sont-ils dangereusement similaires à leur source ? Analyse par plongements de phrases.Actes de CORIA-TALN 2023. Actes de la 30e Conférence sur le Traitement Automatique des Langues Naturelles (TALN), volume 2 : travaux de recherche originaux -- articles courts18e Conférence en Recherche d'Information et Applications -- 16e Rencontres Jeunes Chercheurs en RI -- 30e Conférence sur le Traitement Automatique des Langues Naturelles -- 25e Rencontre des Étudiants Chercheurs en Informatique pour le Traitement Automatique des LanguesParis, FranceATALA2023, 46-54HALback to text
• 30 inproceedingsN.Nicolas Hiebel, O.Olivier Ferret, K.Karën Fort and A.Aurélie Névéol. Similarité surfacique et similarité sémantique dans des cas cliniques générés.Journée d'étude sur la Similarité entre Patients, ATALA, SimPa 2023Paris, FranceMarch 2023HAL
• 31 inproceedingsS.Sylvain Kahane, S.Santiago Herrera, B.Bruno Guillaume and K.Kim Gerdes. Autogramm : développement simultané de treebanks et de grammaires à partir de corpus.Actes de CORIA-TALN 2023. Actes de la 30e Conférence sur le Traitement Automatique des Langues Naturelles (TALN), volume 6 : projets18e Conférence en Recherche d'Information et Applications, 16e Rencontres Jeunes Chercheurs en RI, 30e Conférence sur le Traitement Automatique des Langues Naturelles, 25e Rencontre des Étudiants Chercheurs en Informatique pour le Traitement Automatique des LanguesParis, FranceATALA2023, 37-42HALback to text
• 32 inproceedingsS.Siyana Pavlova, M.Maxime Amblard and B.Bruno Guillaume. A Layered Approach to Semantic Representation.Journées scientifiques du GDR Lift - LIFT 2023Nancy, FranceNovember 2023HALback to text
• 33 inproceedingsV. D.Valentin D. Richard. Can French Interrogative Retrieval be Fully Machine-Based ?Actes des journées LIFT 20235èmes journées du Groupement de Recherche CNRS « Linguistique Informatique, Formelle et de Terrain » (LIFT 2023)Nancy, FranceNovember 2023, 69-76HALback to text

Conferences without proceedings

• 34 inproceedingsM. M.Marc M Anderson. AI Ethics and the Lessons of History.2nd International Conference on the Ethics of Artificial IntelligencePorto, PortugalNovember 2023HALback to text
• 35 inproceedingsM. M.Marc M Anderson. AI as Philosophical Ideology: A Critical look back at McCarthy’s Program.Philosophy in Technology Workshop 2nd EditionWrocław, PolandApril 2023HAL
• 36 inproceedingsF.Fanny Ducel, A.Aurélie Névéol and K.Karën Fort. Bias Identification in Language Models is Biased.Workshop on Algorithmic Injustice 2023Amsterdam, NetherlandsJune 2023HALback to text
• 37 inproceedingsP.Philippe de Groote, M.Maxime Guillaume, A.Agathe Helman, S.Sylvain Pogodalla and R.Raphaël Salmon. Extending Abstract Categorial Grammars with Feature Structures: Theory and Practice.Logic and Engineering of Natural Language Semantics 20 (LENLS20)Osaka, JapanNovember 2023HALback to text
• 38 inproceedingsP.Philippe de Groote. On the semantics of dependencies: relative clauses and open clausal complements - extended abstract -.Logic and Engineering of Natural Language Semantics 20 (LENLS20)Osaka, JapanNovember 2023HALback to text
• 39 inproceedingsJ.Jacques Jayez. (Innocent?) Bias in Argumentation.IMPAQTS (Implicit Manipulation in Politics – Quantitatively Assessing the Tendentiousness of Speeches) final conferenceRome, ItalyApril 2023HALback to text
• 40 inproceedingsA.Agata Savary, C.Cherifa Ben Khelil, C.Carlos Ramisch, V.Voula Giouli, V.Verginica Barbu Mititelu, N.Najet Hadj Mohamed, C.Cvetana Krstev, C.Chaya Liebeskind, H.Hongzhi Xu, S.Sara Stymne, T.Tunga Güngör, T.Thomas Pickard, B.Bruno Guillaume, A.Archna Bhatia, M.Marie Candito, P.Polona Gantar, U.Uxoa Iñurrieta, A.Albert Gatt, K.Kovalevskaite Jolanta, K.Krek Simon, T.Timm Lichte, N.Nikola Ljubešić, J.Johanna Monti, C.Carla Parra Escartín, M.Mehrnoush Shamsfard, I.Ivelina Stoyanova, V.Veronika Vincze and A.Abigail Walsh. PARSEME corpus release 1.3.Proceedings of the 19th Workshop on Multiword Expressions (MWE 2023)Dubrovnik, SloveniaAssociation for Computational LinguisticsMay 2023, 24-35HALDOIback to text

Scientific book chapters

• 41 inbookK.Kim Gerdes, B.Bruno Guillaume, S.Sylvain Kahane and G.Guy Perrier. Function words in Surface-Syntactic Universal Dependencies.Function words in dependency syntaxJohn Benjamins Publishing Company2023HALback to text
• 42 inbookP.Philippe de Groote. Deriving Formal Semantic Representations from Dependency Structures.14213Logic and Engineering of Natural Language Semantics,19th International Conference, LENLS19, Tokyo, Japan, November 19–21, 2022, Revised Selected PapersLecture Notes in Computer ScienceSpringerOctober 2023, 157-172HALDOIback to text

Edition (books, proceedings, special issue of a journal)

• 43 proceedingsM. M.Marc M AndersonExploring the Idea of Ethical Sustainability for Digital Manufacturing.Service Oriented, Holonic and Multi-Agent Manufacturing Systems for Industry of the Future. SOHOMA 2023. Studies in Computational Intelligence, Springer2023HALback to text
• 44 proceedingsK.Karën FortC.Claire GardentY.Yannick ParmentierActes des 5èmes journées du Groupement de Recherche CNRS « Linguistique Informatique, Formelle et de Terrain ».November 2023, 135HALback to text
• 45 periodicalLambek–Grishin Calculus: Focusing, Display and Full Polarization.Outstanding Contributions to Logic25Samson Abramsky on Logic and Structure in Computer Science and BeyondAugust 2023, 877-915HALDOI
• 46 proceedingsV. D.Valentin D. RichardF.Floris RoelofsenProceedings of the 4th Workshop on Inquisitiveness Below and Beyond the Sentence Boundary.4th Workshop on Inquisitiveness Below and Beyond the Sentence Boundary (InqBnB4)Association for Computational Linguistics2023HAL

Doctoral dissertations and habilitation theses

• 47 thesisC.Chuyuan Li. Facing Data Scarcity in Dialogues for Discourse Structure Discovery and Prediction.Université de LorraineAugust 2023HAL

Reports & preprints

• 48 miscJ.Jacques Jayez. Argumentation et probabilités, ou pourquoi l'argumentation rationnelle n'est pas (toujours) un raisonnement.December 2023HALback to text
• 49 reportY.Yannick Parmentier, S.Sylvain Pogodalla, R.Rachel Bawden, M.Matthieu Labeau and I.Iris Eshkol-Taravella. Procédure de diffusion des publications de l'ATALA sur les archives ouvertes.ATALASeptember 2023, 17HAL
• 50 miscV. D.Valentin D. Richard. The costs of Artificial Intelligence: Un aperçu des problèmes éthiques de l'IA.October 2023HALback to text

Other scientific publications

• 51 miscL.Luciana Benotti, K.Karën Fort, M.-Y.Min-Yen Kan and Y.Yulia Tsvetkov. Understanding Ethics in NLP Authoring and Reviewing.Dubrovnic, CroatiaMay 2023, 19-24HALDOIback to text
• 52 inproceedingsB.Bruno Guillaume. Multi-layer querying in Corpora: Example of Parseme and UD.UniDive 1st general meetingSaclay, FranceMarch 2023HAL

Scientific popularization

• 53 miscK.Karën Fort and J.Joanna Jongwane. Quelle éthique pour les agents conversationnels ? [podcast].September 2023HALback to text