Speech Act Modeling of Written Asynchronous Conversations with Task-Specific Embeddings and Conditional Structured Models

Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics, pages 1746–1756, Berlin, Germany, August 7-12, 2016. c⃝2016 Association for Computational Linguistics Speech Act Modeling of Written Asynchronous Conversations with Task-Specific Embeddings and Conditional Structured Models Shafiq Joty and Enamul Hoque ALT Research Group Qatar Computing Research Institute — HBKU, Qatar Foundation {sjoty, mprince}@qf.org.qa Abstract This paper addresses the problem of speech act recognition in written asyn- chronous conversations (e.g., fora, emails). We propose a class of conditional structured models defined over arbitrary graph structures to capture the conversa- tional dependencies between sentences. Our models use sentence representations encoded by a long short term memory (LSTM) recurrent neural model. Empir- ical evaluation shows the effectiveness of our approach over existing ones: (i) LSTMs provide better task-specific representations, and (ii) the global joint model improves over local models. 1 Introduction Asynchronous conversations, where participants communicate with each other at different times (e.g., fora, emails), have become very common for discussing events, issues, queries and life experi- ences. In doing so, participants interact with each other in complex ways, performing certain com- municative acts like asking questions, requesting information or suggesting something. These are called speech acts (Austin, 1962). For example, consider the excerpt of a forum conversation from our corpus in Figure 1. The participant who posted the first comment C1, de- scribes his situation by the first two sentences and then asks a question in the third sentence. Other participants respond to the query by suggesting something or asking for clarification. In this pro- cess, the participants get into a conversation by taking turns, each of which consists of one or more speech acts. The two-part structures across posts like ‘question-answer’ and ‘request-grant’ are called adjacency pairs (Schegloff, 1968). C1: My son wish to do his bachelor degree in Mechanical Engineering in an affordable Canadian university. Human: st, Local: st, Global: st The info. available in the net and the people who wish to offer services are too many and some are misleading. Human: st, Local: st, Global: st The preliminary preparations,eligibility,the require funds etc., are some of the issues which I wish to know from any panel members of this forum .. (truncated) Human: ques, Local: st, Global: st C3 (truncated)...take a list of canadian universities and then create a table and insert all the relevant information by reading each and every program info on the web. Human: sug, Local: sug, Global: sug Without doing a research my advice would be to apply to UVIC .. for the following reasons .. (truncated) Human: sug, Local: sug, Global: sug UBC is good too... but it is expensive particularly for international students due to tuition .. (truncated) Human: sug, Local: sug, Global: sug most of them accept on-line or email application. Human: st, Local: st, Global: st Good luck !! Human: pol, Local: pol, Global: pol C4 snakyy21: UVIC is a short form of? I have already started researching for my brother and found “College of North Atlantic” and …

Content truncated to 500 words