In this paper, we describe the Template Matcher, a system built at SRI to provide robust natural-language interpretation in the Air Travel Information System (ATIS) domain. The system appears to be robust to both speech recognition errors and unanticipated or difficult locutions used by speakers. We explain the motivation for the Template Matcher, describe in general terms how it works in comparison with similar systems, and examine its performance. We discuss some limitations of this approach, and sketch a plan for integrating the Template Matcher with an analytic parser, which we believe will combine the advantages of both.

NLP researchers face a dilemma: on one side, it is unarguably accepted that languages have internal structure rather than strings of words. On the other side, they find it very difficult and expensive to write grammars that have good coverage of language structures. Statistical machine translation tries to cope with this problem by ignoring language structures and using a statistical models to depict the translation process. Most of the translation models are word-based. While the approach has achieved surprisingly good performance comparable to the best commercial systems, many questions remain in the machine translation community. Can the statistical word-based translation still perform well on language pairs with radically different linguistic structures? How would it function with less training data or with spoken languages? The thesis work investigated these questions. In summary, word-based alignment model is a major cause of errors in German-English statistical spoken language...

This paper describes a robust parsing algorithm for spoken language understanding. Comparing with the other work in robust parsing, we focus on building a parser that is robust to not only ill-formed spontaneous spoken language inputs but also under-specified grammars. Preliminary experiment results show that the parsing performance deteriorates more gracefully than another parser we have used when the grammar is more underspecified. 1.

. The Janus-III translation system translates spoken languages in limited domains. The current research focus is on is expanding beyond tasks involving a single semantic domain. The system combines translation components from multiple semantic domains into a unified system using multi-domain parse lattices. This approach yields solutions to several problems including ambiguity resolution, segmentation of spoken utterances into sentence units, modularity of system design, and re-use of earlier systems with incompatible output. Keywords: speech translation, robust parsing, semantic grammars, multi domain, SOUP, JRTk 1. Introduction Spoken Language Translation (SLT) systems have broken many barriers in the 1990's. Translation of well-formed, read speech with a small vocabulary has been replaced with translation of possibly ill-formed, spontaneous speech with a large vocabulary. A remaining limitation for SLT is that it is usually confined to a particular semantic domain. In this paper ...

As part of the JANUS speech-to-speech translation project[5], we have developed a translation system that successfully parses full utterances and is effective in parsing spontaneous speech, which is often syntactically ill-formed. The system is concept-based, meaning that it has no explicit notion of a sentence but rather views each input utterance as a potential sequence of concepts. Generation is performed by translating each of these concepts in whole phrases into the target language, consulting lookup tables only for low-level concepts such as numbers. Currently, we are working on an appointment scheduling task, parsing English, German, Spanish, and Korean input and producing output in those same languages and also Japanese. 1

As part of the JANUS speech-to-speech translation project, we have developed a robust translation system based on the information structures inherent to the task being performed. The basic premise is that the structure of the information to be transmitted is largely independent of the language used to encode it. Our system performs no syntactic analysis; speaker utterances are parsed into semantic chunks, which can be strung together without grammatical rules, and passed through a simple template-based translation module. We have achieved encouraging coverage rates on English, German and Spanish input with English, German and Spanish output.

The MT engine of the Janus speech-to-speech translation system is designed around four main principles: 1) an interlingua approach that allows the e cient addition of new languages, 2) the use of semantic grammars that yield low cost high quality translations for limited domains, 3) modular grammars that support easy expansion into new domains, and 4) e cient integration of multiple grammars using multi-domain parse lattices and domain re-scoring. Within the framework of the C-STAR-II speech-to-speech translation effort, these principles are tested against the challenge of providing translation for a number of domains and language pairs with the additional restriction of a common interchange format.

Low-proficiency non-native speakers represent a significant challenge for large-vocabulary continuous speech recognition (LVCSR). Acoustic models are confused by a heavy accent

The focus of this thesis proposal is to improve the ability of a computational system to understand spoken utterances in a dialogue with a human. Available computational methods for word recognition do not perform as well on spontaneous speech as we would hope. Even a state of the art recognizer achieves slightly worse than 70% word accuracy on (nearly) spontaneous speech in a conversation about a specific problem. To address this problem, I will explore novel methods for post-processing the output of a speech recognizer in order to correct errors. I adopt statistical techniques for modeling the noisy channel from the speaker to the listener in order to correct some of the errors introduced there. The statistical model accounts for frequent errors such as simple word/word confusions and short phrasal problems (one-to-many word substitutionsand many-to-one word concatenations). To use the model, a search algorithm is required to find the most likely correction of a given word sequence ...

The Air Travel Information Service (ATIS) is the designated common task of the ARPA Spoken Language Systems Program. The specified task is to build and evaluate a system capable of handling continuous and spontaneous speech recognition as well as natural language understanding in the ATIS domain. The goal of this research is to develop an effective natural language component for the complete system, to answer queries posed through text input instead of speech. We limit our scope to deal only with those sentences which can be understood unambiguously out of context (the so-called "Class A" queries). Specifically, we wish to use the training data to assign a probability distribution to the reference interpretation, the NLParse, which will minimize the observed perplexity of our test data. The decoder component of the finished system will use the natural language probabilities to select the most probable NLParse translations for a given English input. The NLParse translation can then be unambiguously converted to SQL to find the correct answer.