Identifying sound descriptions in Written Documents
Suzanne Mpouli, Christine Largeron, Michel Beigbeder · 2019
Automatically identifying sound descriptions can be useful to extract soundscapes from written documents. In this respect, we present a preliminary work on the identification of sound descriptions which first expands a list of sound terms taken from a machine-readable lexical resource. Then, to find out the producer of each sound mentioned in a text collection, two methods have been tested out: one which relies on dependency parsing and the other which uses a pre-trained word embedding model. Both methods have been evaluated on a set of sentences extracted from three 19th century French novels; the obtained results suggest that dependency parsing performs better on this specific task. It should be noted that the first step of the proposed system is domain-independent and can generate a domain-specific lexicon for any topic.