{ "paper_id": "J03-1008", "header": { "generated_with": "S2ORC 1.0.0", "date_generated": "2023-01-19T02:15:27.673397Z" }, "title": "", "authors": [], "year": "", "venue": null, "identifiers": {}, "abstract": "", "pdf_parse": { "paper_id": "J03-1008", "_pdf_hash": "", "abstract": [], "body_text": [ { "text": "Manny Rayner, David Carter, Pierrette Bouillon, Vassilis Digalakis, and Mats Wir\u00e9n (editors) (SRI International, Speech Machines, University of Geneva, Technical University of Crete, and Telia Research) Cambridge University Press (Studies in natural language processing, edited by Branimir Boguraev), 2000, xviii+337 pp; hardbound, ISBN 0-521-77077-7, $59.95 The Spoken Language Translator consists of 21 chapters contributed by different authors who worked on the building of components and/or the overall system for the Spoken Language Translator (SLT), an early project in the area of automatic speech translation. The book presents a detailed description of various technologies developed during the period of the SLT project, with an emphasis on language and linguistic processing. It includes four parts: language processing and corpora, linguistic coverage, speech processing, and evaluation and conclusions.", "cite_spans": [ { "start": 176, "end": 202, "text": "Crete, and Telia Research)", "ref_id": null }, { "start": 301, "end": 320, "text": "2000, xviii+337 pp;", "ref_id": null }, { "start": 321, "end": 358, "text": "hardbound, ISBN 0-521-77077-7, $59.95", "ref_id": null } ], "ref_spans": [], "eq_spans": [], "section": "The Spoken Language Translator", "sec_num": null }, { "text": "The SLT used a multiengine machine translation architecture. The main machine translation engine is a complex, unification grammar-based system intended to perform deep analysis and to produce high-quality output. The authors claim that the handcoded grammars used in SLT are linguistically motivated and generic in nature. Therefore, the grammar coverage is broad, and it is feasible to use these methods in serious applications. However, the translation task was evaluated only in the domain of air travel reservations. The authors also worked on methods to convert grammars generated for English to corresponding grammars for French and Swedish, as well as the possibility of writing a single large grammar for two closely related languages. The grammarbased subsystem is responsible for structural translation via the quasi-logical form transfer method.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "The Spoken Language Translator", "sec_num": null }, { "text": "Another translation engine used in SLT is a simple word-to-word system, intended for shallow processing, speed, robustness, and broad coverage. The word-to-word translation component simply uses rules to associate a source language phrase of one or more words with a target language phase of zero or more words. Source language words may be tagged with part-of-speech labels. This technique was intended to fill in the gaps (i.e., to translate utterances that are not covered by the grammars, including utterances that are grammatically incorrect, such as speech recognition errors or disfluencies in spontaneous speech).", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "The Spoken Language Translator", "sec_num": null }, { "text": "The speech recognition technology used in SLT was a state-of-the-art hidden Markov model-based approach, with some interesting new techniques, such as discrete-mixture models, which run two to three times faster than continuous-density models with similar recognition performance. The speech recognition system was developed for multilingual speech and is capable of decoding a word string in any of a given set of languages. Both acoustic modeling and language modeling issues for a multilingual recognition system are addressed. Text-tospeech translation is an inevitable part of a speech-to-speech translation system. However, text-to-speech technology is not considered in this book, and prosody translation and prosody transfer are described only briefly.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "The Spoken Language Translator", "sec_num": null }, { "text": "A series of empirical evaluation experiments is thoroughly described and the results analyzed in chapter 20. Although all evaluations are subjective, several fine scales are used for quality judgment. Furthermore, the main system performance is compared when variations are generated either by replacing selected components or changing the values of the critical parameters. One interesting study discussed in this chapter is on \"pipeline synergy,\" which is based on the intuition that \"utterances that are hard to hear are also hard to understand and translate.\" The authors show that assuming that error rates for the individual components are independent can lead to serious overestimates of the system error rate. This chapter is worthwhile for readers who are interested in performance evaluation of speech-to-speech translation systems.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "The Spoken Language Translator", "sec_num": null }, { "text": "In general, this book is a useful resource for those who are interested in knowing how to build a speech-to-speech translation system, especially for those who are interested in grammar-based language processing approaches. \"The title of this collection of articles, Many Morphologies, is an allusion not only to the variety of morphological problems discussed, but also to the diversity of formal solutions offered by the authors. The papers examine Polish derivational morphology, French and English compounding, pluralization in Luise\u00f1o and Somali, and more, and though almost all of the papers are written in the generative grammar framework, there are considerable differences between the formal methods they apply to the problems at hand. We also include two papers which discuss computational morphology, one an overview of the current state of the field and the other an in-depth study of a current research project.\"-From the introduction", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "The Spoken Language Translator", "sec_num": null }, { "text": "Lars Borin (editor) (Uppsala University) Amsterdam: Editions Rodopi (Language and computers: Studies in practical linguistics 43, edited by Jan Aarts and Willem Meijs), 2002, vii+220 pp; hardbound, ISBN 90-420-1530-6, $50.00, C --50.00", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Parallel Corpora, Parallel Worlds", "sec_num": null }, { "text": "The volume contains selected papers from a symposium on parallel and comparable corpora that was held at Uppsala University in April 1999: \". . . And never the twain shall meet?\" by Lars Borin \"Towards a multilingual corpus for contrastive analysis and translation studies\" by Stig Johansson \"The PLUG project: Parallel corpora in Link\u00f6ping, Uppsala, Goteb\u00f6rg: Aims and achievements\" by Anna S\u00e5gvall Hein \"The Uppsala Student English Corpus (USE): A multi-faceted resource for research and course development\" by Margareta Westergren Axelsson and Ylva Berglund \"How can linguists profit from parallel corpora?\" by Raphael Salkie \"Parallel corpora as tools for investigating and developing minority languages\" by Trond Trosterud \"Reversing a Swedish-English dictionary for the Internet\" by Christer Geisler \"Multilingual corpus-based extraction and the Very Large Lexicon\" by Gregory Grefenstette \"The PLUG Link Annotator-Interactive construction of data from parallel corpora\" by Magnus Merkel, Mikael Andersson, and Lars Ahrenberg \"Building and processing a multilingual corpus of parallel texts\" by Peter Stahl \"Uplug-A modular corpus tool for parallel corpora\" by J\u00f6rg Tiedemann \"Part-of-speech tagging for Swedish\" by The volume contains selected papers from the 12th Computational Linguistics in the Netherlands meeting, Enschede, November 2001: \"Ideas on multi-layer dialogue management for multi-party, multi-conversation, multi-modal communication\" by David R. Traum \"The Alpino dependency treebank\" by Leonoor van der Beek, Gosse Bouma, Rob Malouf, and Gertjan van Noord \"Corpus-based acquisition of collocational prepositional phrases\" by Gosse Bouma and Bego\u00f1a Villada", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Parallel Corpora, Parallel Worlds", "sec_num": null } ], "back_matter": [], "bib_entries": { "BIBREF0": { "ref_id": "b0", "title": "Incremental generation of self-corrections using underspecification\" by Markus Guhe and Frank Schilder \"Creating a Dutch information retrieval test corpus\" by Djoerd Hiemstra and David van Leeuwen \"Performance grammar: A declarative definition", "authors": [ { "first": "Bart", "middle": [], "last": "Decadt", "suffix": "" }, { "first": "Jacques", "middle": [], "last": "Duchateau", "suffix": "" }, { "first": "Walter", "middle": [], "last": "Daelemans", "suffix": "" }, { "first": "Patrick", "middle": [], "last": "Wambacq ; Fien De", "suffix": "" }, { "first": "Walter", "middle": [], "last": "Meulder", "suffix": "" }, { "first": "V\u00e9ronique", "middle": [], "last": "Daelemans", "suffix": "" }, { "first": "", "middle": [], "last": "Hoste", "suffix": "" } ], "year": null, "venue": "Conservative vs set-driven learning functions for the classes k-valued\" by Christophe Costa Flor\u00eancio", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Conservative vs set-driven learning functions for the classes k-valued\" by Christophe Costa Flor\u00eancio \"Memory-based phoneme-to-grapheme conversion\" by Bart Decadt, Jacques Duchateau, Walter Daelemans, and Patrick Wambacq \"Tagging the Dutch PAROLE corpus\" by Jesse de Does and John van der Voort van der Kleij \"A named entity recognition system for Dutch\" by Fien De Meulder, Walter Daelemans, and V\u00e9ronique Hoste \"Reference resolution in context\" by Jan van Eijck \"Accurate stemming of Dutch for text classification\" by Tanja Gaustad and Gosse Bouma \"Incremental generation of self-corrections using underspecification\" by Markus Guhe and Frank Schilder \"Creating a Dutch information retrieval test corpus\" by Djoerd Hiemstra and David van Leeuwen \"Performance grammar: A declarative definition\" by Gerard Kempen and Karin Harbusch \"Multi-feature error detection in spoken dialogue systems\" by Piroska Lendvai, Antal van den Bosch, Emiel Krahmer, and Marc Swerts \"Applying Monte Carlo techniques to language identification\" by Arjen Poutsma \"Automated compounding as a means for maximizing lexical coverage\" by Vincent Vandeghinste", "links": null } }, "ref_entries": { "FIGREF0": { "num": null, "uris": null, "type_str": "figure", "text": "Rodopi (Language and computers: Studies in practical linguistics 45, edited by Jan Aarts and Willem Meijs), 2002, viii+207 pp; hardbound, ISBN 90-420-0943-8, $50.00, C --50.00" }, "TABREF0": { "type_str": "table", "num": null, "content": "
Many Morphologies
Paul Boucher (editor)
(University of Nantes)
Somerville, MA: Cascadilla Press, 2002,
xv+267 pp; paperbound, ISBN
1-57473-025-8, $28.95
", "text": "-Yuqing Gao, IBM T. J. Watson Research Center Yuqing Gao works on problems of large-vocabulary speech recognition and of speech-to-speech translation. She has been a research staff member at IBM T. J. Watson Research Center since 1995. Her address is IBM T. J. Watson Research Center, Yorktown Heights, NY 10598, e-mail: yuqing@us.ibm.com.", "html": null } } } }