Seminar Nicole Novielli

12 Jun 2013 - 15:00 to 17:00

Exploring the lexical semantics of dialogue acts: Lexical profiles of communicative acts, affect analysis and HMM modeling of user engagement
Nicole Novielli, Università degli Studi di Bari ‘Aldo Moro’

When engaged in dialogues, people ask for information, agree with their partner, state some facts and express opinions. They proceed in their conversations through a series of dialogue acts to yield some particular communicative intention. 
Dialogue Acts (DA) are well studied in linguistics and computational linguistics research since long time and can be identified with the communicative goal of an utterance at the level of its illocutionary force. There is a large number of application domains that can benefit from the automatic extraction of the underlying structure of dialogues: dialogue systems for human-computer interaction, conversational agents for monitoring and supporting human-human conversations, analysis of blogs, forums and chat logs for opinion mining, automatic meeting summarization, and so on. Such applications require a deep understanding of the conversational structure: at every step of the interaction the system should be aware of who is telling what to whom. Traditionally, the problem of DA recognition has been addressed with promising results using approaches developed in supervised frameworks. Though, it is not always easy to have large training material at disposal, partly because of manual labeling effort. Moreover, with the advent of the Web, a large amount of material about unstructured and unlabeled natural language interactions (e.g. on blogs and online social networks) has become available, raising the attractiveness of empirical methods analysis on this field. In this talk I will present an unsupervised method for DA recognition that simply exploit lexical semantics of the sentences. Moreover, I will discuss the importance of investigating the role of affective lexicon in the disambiguation of the communicative goal of an utterance. To conclude, I will describe a practical application of DA coding as part of an approach based on Hidden Markov Models, for modeling the engagement experienced by the users interacting with an Embodied Conversational Agent.