| 1. |
- Becket, Ralph, et al.
(author)
-
Spoken Language Translator: Phase Two Report
- 1997
-
Reports (other academic/artistic)abstract
- Spoken Language Translator (SLT) is a project whose long-term goal is the construction of practically useful systems capable of translating human speech from one language into another. The current SLT prototype, described in detail in this report, is capable of speech-to-speech translation between English and Swedish in either direction within the domain of airline flight inquiries, using a vocabulary of about 1500 words. Translation from English and Swedish into French is also possible, with slightly poorer performance.A good English-language speech recognizer existed before the start of the project, and has since been improved in several ways. During the project, we have constructed a Swedish-language recognizer, arguably the best system of its kind so far built. This has involved among other things collection of a large amount of Swedish training data. The recognizer is essentially domain-independent, but has been tuned to give high performance in the air travel inquiry domain.The main version of the Swedish recognizer is trained on the Stockholm dialect of Swedish, and achieves near-real-time performance with a word error rate of about 7%. Techniques developed partly under this project make it possible to port the recognizer to other Swedish dialects using only modest quantities of training data.On the language-processing side, we had at the start of the project a substantial domain-independent language-processing system for English, a preliminary Swedish version, and a sketchy set of rules to permit English to Swedish translation. We now have good versions of the language-processing system for English, Swedish and French, and fair to good support for translation in five of the six possible language- pairs. Translation is carried out using a novel robust architecture developed under the project. In essence, this translates as much of the input utterance as possible using a sophisticated grammar-based method, and then employs a much simpler set of word- to-word translation rules to fill in the gaps.The language-processing modules are all generic in nature, are based on large, linguistically motivated grammars, and can fairly easily be tuned to give good performance in new domains. Much of the work involved in the domain adaptation process can be carried out by non-experts using tools developed under the project.Formal comparisons are problematic, in view of the different domains and languages used and the lack of accepted evaluation criteria. None the less, the evidence at our disposal suggests that the current SLT prototype is no worse than the German Verbmobil demonstrator, in spite of a difference in project budget of more than an order of magnitude.
|
|
| 2. |
- Carter, David, et al.
(author)
-
Common speech/language issues
- 2000
-
In: The spoken language translator. - Cambridge : Cambridge University Press. - 0521770777 - 9780521038829 ; , s. 284-294
-
Book chapter (other academic/artistic)
|
|
| 3. |
- Carter, David, et al.
(author)
-
Common Speech–Language Issues
- 2000
-
In: The Spoken Language Translator. - Cambridge : Cambridge University Press. - 0521770777 - 9780521038829 ; , s. 284-294
-
Book chapter (peer-reviewed)abstract
- This book presents a detailed description of Spoken Language Translator (SLT), one of the first major projects in the area of automatic speech translation. The SLT system can translate between English, French, and Swedish in the domain of air travel planning, using a vocabulary of about 1500 words, and with an accuracy of about 75%. The greater part of the book describes the language processing components, which are largely built on top of the SRI Core Language Engine, using a combination of general grammars and techniques that allow them to be rapidly customized to specific domains. Speech recognition is based on Hidden Markov Mode technology, and uses versions of the SRI DECIPHER system. This account of the Spoken Language Translator should be an essential resource both for those who wish to know what is achievable in spoken-language translation today, and for those who wish to understand how to achieve it.
|
|
| 4. |
- Eklund, Robert, 1962-, et al.
(author)
-
Inclusion of a prosodic module in spoken language translation
- 1995
-
In: Journal of the Acoustical Society of America. - : American Institute of Physics (AIP). - 0001-4966 .- 1520-8524. ; 98:5, s. 2894-2895
-
Journal article (peer-reviewed)abstract
- Current speech recognition systems mainly work on statistical bases and make no use of information signalled by prosody, i.e. the segment duration and fundamental frequency contour of the speech signal. In more advanced applications for speech recognition, such as speech-to-speech translation systems, it is necessary to include the linguistic information conveyed by prosody. Earlier research has shown that prosody conveys information at syntactic, semantic and pragmatic levels. The degree of linguistic information conveyed by prosody varies between languages, from languages such as English, with a relatively low degree of prosodic disambiguation, via tone-accent languages such as Swedish, to pure tone languages. The inclusion of a prosodic module in speech translation systems is not only vital in order to link the source language to the target language, but could also be used to enhance speech recognition proper. Besides syntactic and semantic information, properties such as dialect, sociolect, sex and attitude etc is signalled by prosody. Speech-to-speech recognition systems that will not transfer this type of information will be of limited value for person-to-person communication. A tentative architecture for the inclusion of a prosodic module in a speech-to-speech translation system is presented.
|
|
| 5. |
|
|
| 6. |
- Lyberg, Bertil, et al.
(author)
-
The Possible Use of Prosody in Spoken Language Translation Systems
- 1995
-
In: TELECOM 95. ; , s. 9-13
-
Conference paper (peer-reviewed)abstract
- Speech recognition systems do not normally make use of information signalled by prosody, i.e. the segment duration and the fundamental frequency contour of the speech signal. Rather, in current statistical approaches to the speech recognition problem, the acoustic manifestations of prosody is more or less considered as disturbances. In more advanced applications for speech recognition, such as speech-to-speech translation systems, it is obvious that the information conveyed by prosody has to be detected in the source language, mapped onto the target language and then generated by the speech synthesizer of the target language. The linguistic information signalled by prosody is syntactic structure, semantic interpretation and sentence emphasis. Moreover, in languages such as Swedish, with tonal accents, there are word and phrase pairs that are only distinguishable by means of intonation contour. In pure tone languages, the inclusion of prosody is crucial for speech recognition systems. Besides syntactic and semantic information, prosody also mirrors para-linguistic properties such as sex and attitude etc. Speech-to-speech translation systems that will not transfer this type of information will be of limited value for person-to-person communication.
|
|
