Semi-automatically Alignment of Predicates between Speech and OntoNotes data

Semi-automatically Alignment of Predicates between Speech and OntoNotes Data Niraj Shrestha, Marie-Francine Moens Department of Computer Science, KU Leuven, Belgium {niraj.shrestha, Marie-Francine.Moens}@cs.kuleuven.be Abstract Speech data currently receives a growing attention and is an important source of information. We still lack suitable corpora of transcribed speech annotated with semantic roles that can be used for semantic role labeling (SRL), which is not the case for written data. Semantic role labeling in speech data is a challenging and complex task due to the lack of sentence boundaries and the many transcription errors such as insertion, deletion and misspellings of words. In written data, SRL evaluation is performed at the sentence level, but in speech data sentence boundaries identification is still a bottleneck which makes evaluation more complex. In this work, we semi-automatically align the predicates found in transcribed speech obtained with an automatic speech recognizer (ASR) with the predicates found in the corresponding written documents of the OntoNotes corpus and manually align the semantic roles of these predicates thus obtaining annotated semantic frames in the speech data. This data can serve as gold standard alignments for future research in semantic role labeling of speech data. Keywords: Speech data, Semantic role labeling, Semantics, Automatic speech recognition 1. Introduction Semantic role labeling (SRL) is a process of predicting the predicate-argument structures in language utterances by identifying predicates and their related semantic roles. SRL reveals more information about the content than a syntactic analysis in the field of natural language processing (NLP) in order to better understand “who” did “what” to “whom”, and “how”, “when” and “where”. SRL has many key applications in NLP, such as ques- tion answering, machine translation, and dialogue systems. Many effective SRL systems have been developed to work with written data. However, when applying popular SRL systems such as ASSERT (Pradhan et al., 2005), Lund SRL (Johansson and Nugues, 2008), SWIRL (Surdeanu and Turmo, 2005), and Illinois SRL (Punyakanok et al., 2008) on transcribed speech, which was generated by an automatic speech recognizer (ASR), many errors are made due to the specific nature of the ASR transcribed data. SRL on written data performs well due to the availability of annotated corpora like PropBank (Palmer et al., 2005), FrameNet (Baker et al., 1998) etc., which help to train the SRL system and also the written data is clean and well- formed. Most of the SRL systems on written data are eval- uated at the sentence level. On the other hand, ASR data is noisy and not well-formed, it does not contain sentence boundaries and it exhibits many errors like insertion, dele- tion and misspelling of words. Because of the lack of sen- tence boundaries in speech data and the problem of tran- scribed speech, it is very hard to align speech data and OntoNotes data on the sentence level. Keeping these com- plexities in mind, we align predicates and their semantic roles between speech data and their corresponding writ- ten texts of the OntoNotes corpus, more specifically the OntoNotes release 3.0 …

Content truncated to 500 words