Presentation

---

Spoken Language Processing aims at achieving a robust analysis (syntax, semantics, pragmatics) of spontaneous spoken utterances. It has to cope with two main problems :

• Automatic speech recognition strongly corrupt speech transcriptions,
• Speech disfluences (repairs, self-corrections, hesitations...) break the syntactic structure of spoken language.

I'm investigating robust parsing methods (incremantal shallow parsing based on POS tagging and chunking) in order to overcome these difficulties. This research takes place in the framework of human-machine spoken dialogue or human conversational speech (broadcast speech).

Detection of emotions- I am currently working on the extension of theses researches on the question of the emotion carried by spoken utterances. Most researches conducted in this area are concerning a prosodic characterization of the emotion. In an complementary way, I investigate the detection of emotions from the analysis of the propositionnal content of the spoken utterances. The basic idea is that emotion is compositional : while simple words carry a emotional valency which is established by phycholinguistic works, verbs and adjective act as predicates which take a emotional valency of their arguments to provide a resulting global emotion. As a result, the computation of the general emotion carried by an utterance depends on the semantic structure of the speech turn and lexical emotional norms as well. 

For the moment being, we have implemented a emotion detector (EmoLogus) which is based on the Logus speech understading system . EmoLogus is able to characterise the emotional valency (positive, negative, neutral) and intensity (weak, high) carried by isolated speech turns. We are now working on a more contextual detection of emotion. This work is done in collaboration with the European University of Brittany (Jeanne Villaneau and Dominique Duhaut ) and the Montpellier 3 University (Arielle Syssau-Vaccarella). Agata Savary, from the LI, is also involved in this work.

  Works and projects

---

• Spoken language understaning for dedicated human-machine dialogue - I am developping robust parsing methods which aim at studying tasks that are more complex than standard ATIS-like applications. Furthermore, these approaches describe speech disfluences more precisely than standard pattern-based approches (see for instance the work of Shriberg or Heeman).
  • ROMUS speech understanding system (finite state automata for robust chunking) : PhD of Jerome Goulian (2002) 
  • LOGUS speech understanding system (logical approach based on categorial grammars) : PhD of Jeanne Villaneau (2003) 
• EPAC project (2007-2010) - Extension of our previous works to general conversational speech : chunking and named entities detection of broadcast speech,
• EMOTIROB project (2007-2010) - Speech understanding and detection of émotions for a companion robot that will be used bhy children in hospitals (PhD of Marc Le Tallec under my supervisor and those of Jeanne Villaneau and Dominique Duhaut ).

  Some publications

---

• Marc LE TALLEC, Jeanne VILLANEAU, Jean-Yves ANTOINE, Dominique DUHAUT (2011) Affective Interaction with a Companion Robot for vulnerable Children: a Linguistically based Model for Emotion Detection Proc. LTC’2001, Language Technology Conference, Poznan, Poland. 445-450. [HAL-00664618]

• Marc LE TALLEC, Jeanne VILLANEAU, Jean-Yves ANTOINE, Agata SAVARY, Arielle SYSSAU-VACARELLA A  (2010) Emologus - a compostional model of emotion detection based on the propositionnal content of spoken utterances Proc. 13th International Conference on Text, Speech and Dialogue, TSD'2010, Brno, Czech Republic, sept. 2010 In LNCS/LNAI 6231, Springer, ISBN: 978-3-642-15759-2 [HAL-00536786] .
• Yannick Estève, Thierry Bazillon, Jean-Yves Antoine, Frédéric Béchet, Jérôme Farinas (2010) The EPAC corpus: manual and automatic annotations of conversational speech in French broadcast news. Proc. 9th European conference on Language Resources and Evaluation, LREC’2010, Valetta, Malta, May 2010.  

• Jeanne VILLANEAU, Jean-Yves ANTOINE (2009) Deeper spoken language understanding for man-machine dialogue on broader application domains: a logical alternative to concept spotting. Proc. Workshop on the Semantic Representation of Spoken Language, SRSL’2009, EACL’2009, Athens, Greece, April 2009 . pp. 50-57. [ACM Portal[LREC2010_650] .]

• Jean-Yves ANTOINE, Abdenour MOKRANE, Nathalie FRIBURGER (2008) Automatic rich annotation of large corpus of conversational transcribed speech, Proc. 8th European conference on Language Resources and Evaluation. LREC'2008, Marrakesh, Maroc (à paraître)  [LREC_2008-172] [HAL-00484046]

• Jeanne VILLANEAU, Jean-Yves ANTOINE (2004) Categorial grammars used to partial parsing of spoken language, Proc. Categorial Grammars'2004, Montpellier, France  
• Jerome GOULIAN, Jean-Yves ANTOINE, Franck POIRIER (2003) How NLP techniques can improve speech understanding. ROMUS: a robust chunk based message understanding system using link grammars. Proc.  Eurospeech'2003,  Genève, Suisse, pp. 2773-2776.  
• Jeanne VILLANEAU, Jean-Yves ANTOINE (2001) Combining syntax and pragmatic knowledge for the understanding of spontaneous spoken sentences. Proc. 4th Conference on Logical Aspects of Computational Linguistics, LACL'2001, Le Croisic, France. In P. de Groote, G. Morrill, C. Retore (Eds.) LNAI 2099, Springer Verlag, pp. 279-295.