Pragmatics meets NLP: automatic detection of communicative intentions in Spanish tweets through the Speech Act Theory

The high performance shown by the latest language models created inside the Natural Language Processing (NLP) research area has made these multi-level linguistic approaches to the representation of languages become a scientific hot spot. Nevertheless, those NLP systems arguably tend to set aside the pragmatic level of language due to the added difficulties this linguistic sphere shows, i.e., cultural context, communicative intentions, or implicit information and sarcasm, to name but a few. To boost the inclusion of such parameters in NLP systems to obtain more fine-grained tools with further linguistic knowledge, in this talk we will introduce the annotation scheme we have created following the Speech Act Theory (Austin, 1962; Searle, 1969;1975) to detect the communicative intention of a given message in Spanish from the social network Twitter. Thanks to this annotation scheme, and with the ongoing collaboration with the LISN research group, we aim to find out up to which point intentions can be detected automatically by a language training model, as well as to check how other NLP tasks such as sentiment analysis can contribute to the detection of the intention of a message. By adding such pragmatic information into the system, NLP tools would be able to identify and reproduce the linguistic features that depict the natural essence of language, therefore achieving one of the main objectives of this research area.