What Speakers really Mean when they Ask Questions: Classification of Intentions with a Supervised Approach

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 1159–1166 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 1159 What Speakers really Mean when they Ask Questions : Classification of Intentions with a Supervised Approach Ang`ele Barbedette, Iris Eshkol-Taravella Universit´e Paris Nanterre, MoDyCo UMR 7114 200 avenue de la R´epublique, 92001, Nanterre, France angele.barbedette@gmail.com, ieshkolt@parisnanterre.fr Abstract This paper focuses on the automatic detection of hidden intentions of speakers in questions asked during meals. Our corpus is composed of a set of transcripts of spontaneous oral conversations from ESLO’s corpora. We suggest a typology of these intentions based on our research work and the exploration and annotation of the corpus, in which we define two “explicit” categories (REQUEST FOR AGREEMENT and REQUEST FOR INFORMATION) and three “implicit” categories (OPINION, WILL and DOUBT). We implement a supervised automatic classification model based on annotated data and selected linguistic features and we evaluate its results and performances. We finally try to interpret these results by looking more deeply and specifically into the predictions of the algorithm and the features it used. There are many motivations for this work which are part of ongoing challenges such as opinion analysis, irony detection or the development of conversational agents. Keywords: intention analysis, speech act theory, dialog acts, sentiment analysis, opinion mining 1. Introduction 1.1. Motivations Many of today’s NLP challenges focus on the use of fig- ures of speech to express an idea, an opinion, a will, etc., such as irony (a mocking way of saying something when you think the opposite) or litany (a way of reducing what you say : you say less than what you really think). Most of the existing work on this problem focuses on written or oral non-spontaneous data, which include more easily identifi- able clues to identify and understand these cases, particu- larly through prosody. For this work, we decided to focus only on oral transcriptions, and more specifically on tran- scriptions of recordings of natural and spontaneous conver- sations during meals. In our opinion, these are more chal- lenging because it is less easy to interpret the intentions of the speakers’ utterances in them. The main problem is to be able to determine, without relying on oral and prosodic indicators, what the speaker means in a non-literal way, that is, his intentions, when he asks a question. The aim is to try to meet a NLP need to define and identify the implicit con- tent, which does not only require to find textual clues. This work also provides us with a corpus of annotated natural and spontaneous oral data, a type of data that is difficult to obtain and yet very needed in NLP. There are many possi- ble outcomes to this work like the improvement of human- machine dialogue and chatbots, a deeper opinion analysis and irony detection. 1.2. Organization of the paper We will start by contextualizing this work and defining what is an intention. After presenting our …

Content truncated to 500 words