Finite State Grammar Transduction from Distributed Collected Knowledge

In this paper, we discuss the use of Open Mind Indoor Common Sense (OMICS) project for the purpose of speech recognition of user requests. As part of OMICS data collection, we asked users to enter different ways of asking a robot to perform specific tasks. This paraphrasing data is processed using Natural Language techniques and lexical resources like WordNet to generate a Finite State Grammar Transducer (FSGT). This transducer captures the variations in user requests and captures their structure. We compare the task recognition performance of this FSGT model with an n-gram Statistical Language Model (SLM). The SLM model is trained with the same data that was used to generate the FSGT. The FSGT model and SLM are combined in a two-pass system to optimize full and partial recognition for both in-grammar and out-of-grammar user requests. Our work validates the use of a web based knowledge capture system to harvest phrases to build grammar models. Work was performed using Nuance Speech Recognition system.

[1]  Srinivas Bangalore,et al.  Supertagging: An Approach to Almost Parsing , 1999, CL.

[2]  Douglas E. Appelt,et al.  GEMINI: A Natural Language System for Spoken-Language Understanding , 1993, ACL.

[3]  Alex Acero,et al.  A Semantically Structured Language Model , 2004 .

[4]  Mark Steedman,et al.  Wide-Coverage Semantic Representations from a CCG Parser , 2004, COLING.

[5]  Dan Roth,et al.  Natural Language Inference via Dependency Tree Mapping: An Application to Question Answering , 2004 .

[6]  Paul Martin.  The "Casual Cashmere Diaper Bag": Constraining Speech Recognition Using Examples , 1997, Real Applications@ACL/EACL.

[7]  Xuedong Huang,et al.  A unified context-free grammar and n-gram model for spoken language processing , 2000, 2000 IEEE International Conference on Acoustics, Speech, and Signal Processing. Proceedings (Cat. No.00CH37100).

[8]  T. Landauer,et al.  Indexing by Latent Semantic Analysis , 1990 .

[9]  Eytan Ruppin,et al.  Unsupervised Context Sensitive Language Acquisition from a Large Corpus , 2003, NIPS.

[10]  David G. Stork,et al.  Building intelligent systems one e-citizen at a time , 1999, IEEE Intell. Syst..

[11]  Dan Klein,et al.  Natural language grammar induction with a generative constituent-context model , 2005, Pattern Recognit..

[12]  Dan Klein,et al.  Parsing with Treebank Grammars: Empirical Bounds, Theoretical Models, and the Structure of the Penn Treebank , 2001, ACL.

[13]  Eric Brill,et al.  A Simple Rule-Based Part of Speech Tagger , 1992, HLT.

[14]  Oliver Lemon,et al.  Targeted help for spoken dialogue systems: intelligent feedback improves naive users' performance , 2003 .

[15]  Srinivas Bangalore,et al.  Automatic Acquisition of Phrase Grammars for Stochastic Language Modeling , 1998, VLC@COLING/ACL.

[16]  Dan Klein,et al.  Distributional phrase structure induction , 2001, CoNLL.

[17]  Rakesh Gupta,et al.  Common Sense Data Acquisition for Indoor Mobile Robots , 2004, AAAI.

[18]  Fernando Pereira,et al.  Weighted finite-state transducers in speech recognition , 2002, Comput. Speech Lang..