Learning a semantic parser from spoken utterances

Abstract

Semantic parsers map natural language input into semantic representations. In this paper, we present an approach that learns a semantic parser in the form of a lexicon and an inventory of syntactic patterns from ambiguous training data which is applicable to spoken utterances. We only assume the availability of a task-independent phoneme recognizer, making it easy to adapt to other tasks and yielding no a priori restriction concerning the vocabulary that the parser can process. In spite of these low requirements, we show that our approach can be successfully applied to both spoken and written data.

Semantic parsing of speech using grammars learned with weak supervision

“…The algorithm is also applicable to textual input and has been shown to achieve state-of-the-art performance on written input (cf. Gaspers and Cimiano (2014)). The induced parser is represented in the form of a lexicon and an inventory containing syntactic constructions and thus well-suited to be transformed into a rulebased speech recognition grammar.…”

“…Hence, we explore a 4-fold cross-validation scenario in which for each fold learning is performed using the written ambiguous training data for three games, while the spoken gold standard of the forth game is used for testing, i.e. for performing both speech recognition and subsequent parsing of the resulting ASR transcriptions; spoken data are the same as in Gaspers and Cimiano (2014). For application with the ASR we normalized training data which mainly comprised lowercasing and replacement of numbers in player names, e.g.…”

“…Recently, this task has been extended to consider speech data, both in learning and applying a parser. In particular, the approach of Gaspers and Cimiano (2014) relied on transcriptions made by a task-independent phoneme recognizer as input. For this purpose, NL comments contained in the dataset were read by a speaker.…”

“…For semantic parser induction we applied the algorithm presented in Gaspers and Cimiano (2014), which is mainly designed to work with the output of a phoneme recognizer. The algorithm is also applicable to textual input and has been shown to achieve state-of-the-art performance on written input (cf.…”