Active Learning in Noisy Conditions for Spoken Language Understanding

Proceedings of COLING 2014, the 25th International Conference on Computational Linguistics: Technical Papers, pages 1081–1090, Dublin, Ireland, August 23-29 2014. Active Learning in Noisy Conditions for Spoken Language Understanding Abstract Active learning has proved effective in many fields of natural language processing. However, in the field of spoken language understanding which is always dealing with noise, no complete comparison between different active learning methods has been done. This paper compares the best known active learning methods in noisy conditions for spoken language understanding. Additionally a new method based on Fisher information named as Weighted Gradient Uncertainty (WGU) is proposed. Furthermore, Strict Local Density (SLD) method is proposed based on a new concept of local density and a new technique of utilizing information density measures. Results demonstrate that both proposed methods outperform the best performance of the previous methods in noisy and noise-free conditions with SLD being superior to WGU slightly. 1 Introduction Spoken language understanding (SLU) is currently an emerging field in the intersection of speech processing and natural language processing (Tur and De Mori, 2011). The task of an SLU system is to extract meaning from speech utterances. Example real-world applications are AT&T's How May I Help You? and BBN's Call Director. In the field of SLU, as well as other fields of natural language processing, gathering data is fairly cheap but labeling is quite expensive and time-consuming. Thus, active learning methods apply very well and can greatly reduce costs. This article evaluates different techniques of active learning in the context of statistical SLU to reduce the labeling effort as much as possible. Also, SLU deals with the most amount of noise, in comparison with other fields of NLP, making robustness one of its most important issues (Tur and De Mori, 2011). Therefore, in this article noisy conditions of SLU are explored too. In this paper, we concentrate on statistical approaches for modeling the SLU system. Specifically conditional random fields (Lafferty et al., 2001) are used with a flat semantic frame to represent meaning and to model the SLU system. While there have been a couple of studies on active learning in the context of SLU, they have mostly used only methods in the frameworks of uncertainty sampling (Tur et al., 2003; Jars and Panaget, 2008) and query-by-committee (Gotab et al., 2009). In addition, noisy conditions which are an important aspect of SLU have not been addressed thoroughly. In this paper, performance of various known active learning methods namely uncertainty sampling, query-by-committee, Fisher information ratio (Settles and Craven, 2008) and instability sampling (Zhu This work is licenced under a Creative Commons Attribution 4.0 International License. Page numbers and proceedings footer are added by the organizers. License details: http://creativecommons.org/licenses/by/4.0/. Hossein Hadian Department of Computer Engineering, Sharif University of Technology, Tehran, Iran hadian@ce.sharif.edu Hossein Sameti Department of Computer Engineering, Sharif University of Technology, Tehran, Iran sameti@sharif.edu 1081 and Ma, 2012) are examined and analyzed in noise-free and noisy conditions of SLU. Also a new method for measuring informativeness of instances based …

Content truncated to 500 words