Was That a Question? Automatic Classification of Discourse Meaning in Spanish

20141

Authors

nan

Abstract

AbstractThis paper examines the effectiveness of different feature representations of audio data in accurately classifying discourse meaning in Spanish. The task involves determining whether an utterance is a declarative sentence, an interrogative, an imperative, etc. We explore how pitch contour can be represented for a discourse-meaning classification task, employing three different audio features: MFCCs, Mel-scale spectrograms, and chromagrams. We also determine if utilizing means is more effective in representing the speech signal, given the large number of coefficients produced during the feature extraction process. Finally, we evaluate whether these feature representation techniques are sensitive to speaker information. Our results show that a recurrent neural network architecture in conjunction with all three feature sets yields the best results for the task.

Keywords

nan

Full Content

Proceedings of Recent Advances in Natural Language Processing, pages 132–142 Varna, Sep 4–6, 2023 https://doi.org/10.26615/978-954-452-092-2_015 132 Was That a Question? Automatic Classification of Discourse Meaning in Spanish Santiago Arr´oniz Indiana University sarroniz@indiana.edu Sandra K¨ubler Indiana University skuebler@indiana.edu Abstract This paper examines the effectiveness of differ- ent feature representations of audio data in ac- curately classifying discourse meaning in Span- ish. The task involves determining whether an utterance is a declarative sentence, an interrog- ative, an imperative, etc. We explore how pitch contour can be represented for a discourse- meaning classification task, employing three different audio features: MFCCs, Mel-scale spectrograms, and chromagrams. We also de- termine if utilizing means is more effective in representing the speech signal, given the large number of coefficients produced during the fea- ture extraction process. Finally, we evaluate whether these feature representation techniques are sensitive to speaker information. Our re- sults show that a recurrent neural network ar- chitecture in conjunction with all three feature sets yields the best results for the task. 1 Introduction The aim of this study is to investigate the efficacy of feature representations of audio data in accu- rately classifying discourse meaning in Spanish. The task involves determining whether an utter- ance is a declarative sentence, an interrogative, an imperative, etc. Since there does not seem to be an agreed upon name for this task, we will refer to it as discourse meaning (rather than referring to a broader sense of this term). In human perception, this process involves the comprehension of the relationship of words, phrases, and clauses used in a sentence, as well as their overall contribution to the intended meaning of the sentence. We focus on the prosodic features of different discourse meanings in Spanish. Pitch, or the perceived highness or lowness of a sound, can play a role in distinguishing between differ- ent discourse meanings. For example, declarative sentences typically present a falling pitch contour, indicating that the statement is complete, while in- terrogatives usually have a rising contour, signaling that a question is being asked. In contrast to tonal languages such as Mandarin Chinese, Thai, or Punjabi, which mark the phono- logical contrast of pitch at the lexical level (word level), intonational languages such as Spanish or English mark the phonological contrast of pitch at the utterance level. For Spanish, pitch move- ments are mainly used to signal discourse meaning or to mark focus. The properties that govern pro- duction in intonation are structurally analogous to those that govern lexical tones and morphological paradigms (Ladd, 2008). This means that a declar- ative statement like Mar´ıa viene ’Mar´ıa is coming’ and its interrogative counterpart ¿Mar´ıa viene? ’Is Maria coming?’ differ only in the intonational con- tour with which they are produced, since both are syntactically and lexically identical. 1.1 Research Questions Our study focuses on three main research ques- tions: RQ1: How do we represent intonation as features in discourse-meaning classification for Spanish? RQ2: Do different feature representations convey distinct types of information? RQ3: Are these feature …

Content truncated to 500 words

Labeled Text

AbstractThis paper examines the effectiveness of different feature representations of audio data in accurately classifying discourse meaning in Spanish. The task involves determining whether an utterance is a declarative sentence, an interrogative, an imperative, etc. We explore how pitch contour can be represented for a discourse-meaning classification task, employing three different audio features: MFCCs, Mel-scale spectrograms, and chromagrams. We also determine if utilizing means is more effective in representing the speech signal, given the large number of coefficients produced during the feature extraction process. Finally, we evaluate whether these feature representation techniques are sensitive to speaker information. Our results show that a recurrent neural network architecture in conjunction with all three feature sets yields the best results for the task.

Publication Details

Paper ID:
20141
Year:
2023
Publication:
TIPSTER
Publication Type:
conf
DOI:
nan