What is the Real Intention behind this Question? Dataset Collection and Intention Classification

Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 13606–13622 July 9-14, 2023 ©2023 Association for Computational Linguistics What is the Real Intention behind this Question? Dataset Collection and Intention Classification Maryam Sadat Mirzaei, Kourosh Meshgi & Satoshi Sekine RIKEN Center for Advanced Intelligence Project (AIP) Tokyo, Japan {maryam.mirzaei, kourosh.meshgi, satoshi.sekine}@riken.jp Abstract Asking and answering questions are insepara- ble parts of human social life. The primary purposes of asking questions are to gain knowl- edge or request help which has been the subject of question-answering studies. However, ques- tions can also reflect negative intentions and include implicit offenses, such as highlighting one’s lack of knowledge or bolstering an al- leged superior knowledge, which can lead to conflict in conversations; yet has been scarcely researched. This paper is the first study to in- troduce a dataset (Question Intention Dataset) that includes questions with positive/neutral and negative intentions and the underlying in- tention categories within each group. We fur- ther conduct a meta-analysis to highlight tacit and apparent intents. We also propose a classi- fication method using Transformers augmented by TF-IDF-based features and report the re- sults of several models for classifying the main intention categories. We aim to highlight the importance of taking intentions into account, especially implicit and negative ones, to gain insight into conflict-evoking questions and bet- ter understand human-human communication on the web for NLP applications. 1 Introduction The essence of conversation is to communicate in- tentions; however, the uptake of what has been communicated entails more than merely decoding the words in the message (Galinsky et al., 2005). Many layers underlie a communicative message, and as we interact, we try to decode the surface meaning as well as the tacit aspects (Sperber and Wilson, 2015). We further use established codes to interpret the meanings. For instance, a question is known to be a means of asking for information or requesting help. However, as humans, we also apply meta-knowledge to override the established rules; thus, we may interpret a question as deceit- ful, then consider it as a means to criticize. Our Question Have you been invited to edit this article? Positive/ neutral intention Clarification/Confirmation: Inquiry to disambiguate whether an invitation was involved. Negative intention Putdown/Embarrass: An insulting re- mark to inflict harm as no one has per- mitted the person to conduct any edits. Table 1: A question with multiple perceived intentions. interpretation is a byproduct of multiple factors, including utterance and context, as well as our be- liefs, desire, presuppositions, mental states (Well- man, 1992) and cultural background that leads to several possible interpretations of a single message (Creswell, 1996). In this view, Table 1 shows a question with neutral or negative intentions when different perspectives are taken into account. Studies that focus on the intentions of questions in conversations can be categorized as: those focus- ing on question-answering (Soares and Parreiras, 2020), those related to search engines for accurate information retrieval (Kwiatkowski et al., 2019), and studies …

Content truncated to 500 words