The present paper reports on the advantages of using graph databases in the development of dynamic language models in Spoken Language Understanding applications, such as spoken dialogue systems. First of all, we introduce Neo4J graph databases and, specifically, MultiWordNet-Extended, a graph representing linguistic knowledge. After this first overview, we show how information included in graphs can be used in speech recognition grammars to automatically extend a generic rule structure. This can be the case of linguistic elements, such as synonyms, hypernyms, meronyms and phonological neighbours, which are semantically or structurally related to each other in our mental lexicon. In all the AI based approaches depending on a training process using large and representative corpora, the probability to correctly predict the creativity a speaker can perform in using language and posing questions is lower than expected. Trying to capture most of the possible words and expressions a speaker could use is extremely necessary, but even an empirical, finite collection of cases could not be enough. For this reason, the use of our tool appears as an appealing solution, capable of including many pieces of information. In addition, we used the proposed tool to develop a spoken dialogue system for museums and the preliminary results are shown and discussed in this paper.
Graph databases for designing high-performance speech recognition grammars / DI MARO, Maria; Valentino, Marco; Riccio, Anna; Origlia, Antonio. - (2017). (Intervento presentato al convegno International Conference on Computational Semantics).
Graph databases for designing high-performance speech recognition grammars
Di Maro MariaPrimo
Conceptualization
;Origlia AntonioMethodology
2017
Abstract
The present paper reports on the advantages of using graph databases in the development of dynamic language models in Spoken Language Understanding applications, such as spoken dialogue systems. First of all, we introduce Neo4J graph databases and, specifically, MultiWordNet-Extended, a graph representing linguistic knowledge. After this first overview, we show how information included in graphs can be used in speech recognition grammars to automatically extend a generic rule structure. This can be the case of linguistic elements, such as synonyms, hypernyms, meronyms and phonological neighbours, which are semantically or structurally related to each other in our mental lexicon. In all the AI based approaches depending on a training process using large and representative corpora, the probability to correctly predict the creativity a speaker can perform in using language and posing questions is lower than expected. Trying to capture most of the possible words and expressions a speaker could use is extremely necessary, but even an empirical, finite collection of cases could not be enough. For this reason, the use of our tool appears as an appealing solution, capable of including many pieces of information. In addition, we used the proposed tool to develop a spoken dialogue system for museums and the preliminary results are shown and discussed in this paper.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.