Purely Corpus-based Automatic Conversation Authoring

Purely Corpus-based Automatic Conversation Authoring Guillaume Dubuisson Duplessis1, Vincent Letard2, Anne-Laure Ligozat3, Sophie Rosset1 1LIMSI, CNRS, Universit´e Paris-Saclay, F-91405 Orsay 2LIMSI, CNRS, Univ. Paris-Sud, Universit´e Paris-Saclay, F-91405 Orsay 3LIMSI, CNRS, ENSIIE, Universit´e Paris-Saclay, F-91405 Orsay {gdubuisson, letard, annlor, rosset}@limsi.fr Abstract This paper presents an automatic corpus-based process to author an open-domain conversational strategy usable both in chatterbot systems and as a fallback strategy for out-of-domain human utterances. Our approach is implemented on a corpus of television drama subtitles. This system is used as a chatterbot system to collect a corpus of 41 open-domain textual dialogues with 27 human participants. The general capabilities of the system are studied through objective measures and subjective self-reports in terms of understandability, repetition and coherence of the system responses selected in reaction to human utterances. Subjective evaluations of the collected dialogues are presented with respect to amusement, engagement and enjoyability. The main factors influencing those dimensions in our chatterbot experiment are discussed. Keywords: Example-based dialogue modelling; Open-domain dialogue system; Human-Machine dialogue corpus; Evaluation 1. Introduction The main objective of our work is to design a conversa- tional strategy aiming at the maintenance or increase of the human participation in dialogue, usable both in chatterbot systems (such as Eliza (Weizenbaum, 1966)), or as a fall- back strategy for out-of-domain human utterances in spe- cific dialogue systems. An important property for such a system is to enable a wide variety of possible responses, while keeping a level of control of what is said (with regard to, e.g., the desired language level, the relationship between the human and the system, the situation of interaction). Our approach belongs to the category of example-based di- alogue modelling which aims at using a database of seman- tically indexed dialogue examples to manage dialogue (Lee et al., 2009). However, our purpose is a complete automa- tion of this process from the creation of the database to the conversational management process to avoid the need of a costly and time-consuming human intervention. To this end, we present a data-driven process aiming at the auto- matic authoring of a conversational strategy for a dialogue system interacting with a human participant. The main pur- pose of the conversational strategy is to select an appropri- ate response from a corpus given a human utterance, and adapt it by taking into account the history of dialogue (in- cluding the last utterance from the human). This process relies both on the exploitation of a large and varied cor- pus of Human-Human interactions and on natural language processing tools such as a named entity (NE) recogniser. Our approach is implemented in the context of the Joker project which aims to build a generic intelligent user inter- face providing a multimodal dialogue system with social communication skills including humour and other social behaviours (Devillers et al., 2015). This system is primar- ily involved in entertaining interactions occurring in a so- cial environment (e.g., a cafeteria). A conversational strat- egy could be fruitfully used to provide the system with ju- dicious …

Content truncated to 500 words