A Methodology for Dialogue Act Modeling from Part- of-Speech Annotations of Practical Dialogues in Mexican Spanish

This article proposes a methodology to model dialogue acts (DA) from part-of-speech (POS) annotations in practical dialogues in Mexican Spanish. An example using empirical data is presented to illustrate this methodology. DA and POS annotations were obtained from DIME, a spoken dialogue corpus in which speakers perform tasks in a virtual environment on a collaborative basis. POS structure is a key notion in this methodology. It is the POS tagging sequence for particular instances of utterances. The models are produced as probability tables of pairs of DA annotations and POS structures. Results show that a number of POS structures are more frequently used by speakers to communicate particular DA types. These patterns can be exploited to enhance systems for automatic recognition of DA.