Enhancing The RATP-DECODA Corpus With Linguistic Annotations For Performing A Large Range Of NLP Tasks

Enhancing The RATP-DECODA Corpus With Linguistic Annotations For Performing A Large Range Of NLP Tasks Carole Lailler(1), Anaïs Landeau(1), Frédéric Béchet(2), Yannick Estève(1), Paul Deléglise(1) Laboratoire Informatique de l’Université du Maine (LIUM) University of Le Mans, France Aix Marseille Universite (AMU), CNRS-LIF, France (1) firstname.lastname@lium.univ-lemans.fr (2) firstname.lastname@lif.univ-mrs.fr Abstract In this article, we present the RATP-DECODA Corpus which is composed by a set of 67 hours of speech from telephone conversations of a Customer Care Service (CCS). This corpus is already available on line at http://sldr.org/sldr000847/fr in its first version. However, many enhancements have been made in order to allow the development of automatic techniques to transcript conversations and to capture their meaning. These enhancements fall into two categories: firstly, we have increased the size of the corpus with manual transcriptions from a new operational day; secondly we have added new linguistic annotations to the whole corpus (either manually or through an automatic processing) in order to perform various linguistic tasks from syntactic and semantic parsing to dialog act tagging and dialog summarization. Keywords: Corpus, Linguistic Annotations, NLP tasks 1. Introduction The RATP-DECODA corpus (Bechet et al., 2012) has been acquired in different operational days at the CCS of the Paris transportation center (RATP) for a project funded by the French Research Agency (ANR). The corpus consists of real-life telephone conversations and focuses exclusively on spoken language. These conversations introduce a cus- tomer problem to which the agent attempts to provide so- lutions following a protocol. They have duration variable from 1 to 12 minutes. The number of turns in a conversa- tion and the number of words in a turn are highly variable. The majority of the conversations have more than ten turns. The turns of the customer tend to be longer (> 20 words) than those of the agent and they are more likely to contain out of vocabulary words that are often irrelevant for the res- olution of the problem. This corpus gives an illustration of what the actual interactions between customer and user in an information service. All the conversations of the RATP-DECODA corpus have been manually transcribed. In addition, manual annotators have given a theme to each conversation from a list of 14 labels (13 semantic classes and a NULL class). From these manual annotations, several additional layers of linguistic annotations have been added, such as disfluencies, syntac- tic and semantic parses, discourse structure...These anno- tations have been done either manually, or through an auto- mated process with or without human supervision. This paper recalls all the annotation available on the corpus through a presentation of all the linguistic tasks that can be carried on it. 2. The RATP-DECODA corpus Data mining is an area of research which interest is grow- ing. In addition, the recent increase in call-center and the low cost of storing audio data allowed the recording of oral messages and building of databases very large. These call- centers are a challenge for businesses and can help answer two important applications of the …

Content truncated to 500 words