Transition-Based Disfluency Detection using LSTMs

Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, pages 2785–2794 Copenhagen, Denmark, September 7–11, 2017. c⃝2017 Association for Computational Linguistics Transition-Based Disfluency Detection using LSTMs Shaolei Wang1, Wanxiang Che∗1, Yue Zhang2, Meishan Zhang3 and Ting Liu1 1Center for Social Computing and Information Retrieval, Harbin Institute of Technology, China 2Singapore University of Technology and Design 3School of Computer Science and Technology, Heilongjiang University, China {slwang, car, tliu}@ir.hit.edu.cn yue zhang@sutd.edu.sg, mason.zms@gmail.com Abstract We model the problem of disfluency detec- tion using a transition-based framework, which incrementally constructs and labels the disfluency chunk of input sentences us- ing a set of transition actions without syn- tax information. Compared with sequence labeling methods, it can capture non-local chunk-level features; compared with joint parsing and disfluency detection methods, it is free for noise in syntax. Experiments show that our model achieves state-of-the- art F-score on both the commonly used English Switchboard test set and a set of in-house annotated Chinese data. 1 Introduction Disfluency detection is the task of recognizing non-fluent word sequences in spoken language transcripts (Zayats et al., 2016; Wang et al., 2016; Wu et al., 2015). As shown in Figure 1, stan- dard annotation of disfluency structure (Shriberg, 1994) indicates the reparandum (words that are discarded, or corrected by the following words), the interruption point (+) marking the end of the reparandum, the associated repair, and an optional interregnum after the interruption point (filled pauses, discourse cue words, etc.). Ignoring the interregnum, disfluencies can be categorized into three types: restarts, repetitions, and corrections, based on whether the repair is empty, the same as the reparandum or different, respectively. Table 1 gives a few examples. In- terregnums are easy to detect as they often consist of fixed phrases (e.g. “uh”, “you know”). How- ever, reparandums are more difficult to detect, be- cause they can be in arbitrary form. Most previ- *Email corresponding. I want a flight [ to Boston + {um} to Denver ] RM IM RP Figure 1: Sentence with disfluencies annotated in English Switchboard corpus. RM=Reparandum, IM=Interregnum, RP=Repair. The preceding RM is corrected by the following RP. Type Annotation repair [ I just + I ] enjoy working repair [ we want + {well} in our area we want ] to repetition [it’s + {uh} it’s ] almost like restart [ we would like + ] let’s go to the Table 1: Different types of disfluencies. ous disfluency detection work focuses on detect- ing reparandums. The main challenges of detecting reparandums include that they vary in length, may occur in dif- ferent locations, and are sometimes nested. For example, the longest reparandum in our training set has fifteen words. Hence, it is very important to capture long-range dependencies for disfluency detection. Since there is large parallelism between the reparandum chunk and the following repair chunk (for example, in Figure 1, the reparandum begins with to and ends before another occurrence of to), it is also useful to exploit chunk-level rep- resentation, which explicitly makes use of resulted …

Content truncated to 500 words