Textless Dependency Parsing by Labeled Sequence Prediction
This addresses the problem of lexical knowledge capture in textless spoken language processing for researchers in speech and NLP, though it is incremental in showing limitations of current textless methods.
The paper tackles dependency parsing directly from speech signals without automatic speech recognition, representing dependency trees as labeled sequences. While the cascading method achieved higher overall parsing accuracy, the textless method performed better on instances with important acoustic features.
Traditional spoken language processing involves cascading an automatic speech recognition (ASR) system into text processing models. In contrast, "textless" methods process speech representations without ASR systems, enabling the direct use of acoustic speech features. Although their effectiveness is shown in capturing acoustic features, it is unclear in capturing lexical knowledge. This paper proposes a textless method for dependency parsing, examining its effectiveness and limitations. Our proposed method predicts a dependency tree from a speech signal without transcribing, representing the tree as a labeled sequence. scading method outperforms the textless method in overall parsing accuracy, the latter excels in instances with important acoustic features. Our findings highlight the importance of fusing word-level representations and sentence-level prosody for enhanced parsing performance. The code and models are made publicly available: https://github.com/mynlp/SpeechParser.