"Buenos dias", "buenas noches" -- this was the first words in a foreign language I heard in my life, as a three-year old boy growing up in developing post-war Western Germany, where the first gastarbeiters had arrived from Spain. Fascinated by the strange sounds, I tried to get to know some more languages, the only opportunity being TV courses of English and French -- there was no foreign language education for pre-teen school children in Germany yet in those days. Read more
[ We apologise for the duplicate posting of this announcement ]
EUROPEAN LANGUAGE RESOURCES ASSOCIATION (ELRA)
*** NEW CATALOGUE & NEW RESOURCES ***
The new release of ELRA catalogue (vol2N1) has grown up and currently consists of:
1) Spoken resources: 37 databases in several languages (recordings from microphone, telephone, continuous speech, isolated words, phonetic distionaries, etc.).
2) Written resources: * 14 monolingual and multilingual corpora * 28 monolingual lexica * Around 60 multilingual lexica * A linguistic software platform and grammars development platform
3) Terminological resources: over 360 databases with a wide range of domains and several languages (Catalan, Danish, English, French, German, Italian, Latin, Polish, Portuguese, Spanish, Turkish).
Since our last news on this electronic list, new resources have been negotiated by ELRA and are now available. These are:
SPEECH AND RELATED RESOURCES
ELRA-S0035 Phonolex (BAS/DFKI):
PHONOLEX consists of a simple list of word forms (666,237 inflected words) with a set of features e.g. orthography (German 'Umlauts' in LaTeX format, capital nouns, old German spelling rules), linguistic information (nouns, verbs, etc.), pronunciation and a list of empirical pronunciations.
Language: German Format: ASCII Mark-up: extended SAM-PA (PhonDat-Verbmobil)
All dictionaries contain phonetic transcriptions, with related phoneme lists. The following dictionaries are available (the label basic dictionary refers to the above ELRA-S0036):
Domain Entries Banking French 10,200 Banking German 10,200 Banking Italian 10,200 Banking Spanish 10,200 Radiology German 42,000 (including basic dictionary) Radiology English 16,000 Medical German 130,000 (including basic dictionary) Jurisprudence German 31,000 Jurisprudence German 55,000 (including basic dictionary) Insurance German & English 37,000
A peculiarity of medical dictionaries in German speaking countries has to be taken into consideration: doctors in Germany, Austria and Switzerland may not use the original technical terms in Latin but the Latin word in a spelled manner or a German technical term (see examples below). Medical dictionaries therefore have to contain three different terms.
Technical term Technical term Technical term in Latin in German spelling in German
VoiceMail consists of 17,5 hours of read acoustic speech divided into 9,5 hours of transliterated speech and 8 hours of non-transliterated speech recorded over the digital telephone network (ISDN) with 921 speakers originated from the USA. It contains orthographic transliteration for abou 25,000 utterrances (of 34,912 utterances in total).
Language: American English Standard in use: headerless, one separate transliteration file comprising all utterances of all speakers Sampling rate: 8 kHz Speakers: 377 males and 544 females Size: 17,5 hours Medium: 2 CD-ROM
WRITTEN RESOURCES - MONOLINGUAL LEXICA
ELRA-L0021 Dictionary of French verbs - CORA:
This dictionary contains 25,610 verbs with usage domains, level of language (familiar, popular, literary, Quebec and Swiss terms, etc.), conjugation, auxiliary, verbal adjectives in -able, -ant or -E9, encoded syntactical constructions (subject, direct & indirect object, adverb), sample phrases, synonyms, operators enabling semantic-syntactic classification, encoding of derived forms in -age, -ment, -tion, -oir, -ure, deverbal nouns, base words from which verbs can be derived, a scale of usage ranging from 1 to 6, like those used by commercial dictionaries (basic vocabulary, extended, specialised, etc.). Codes enable automatic production of conjugation forms, derived nouns and adjectives and, if necessary, the production of potential forms.
This dictionary is composed of 126,844 words, with usage domains, grammatical category, gender, number, uncountable, collective, adjectival, nominal, verbal, adverbial derived forms according to the type of words.
4,286 suffixes and prefixes, plus information on their verbal, nominal or adjectival bases or on the verbal basis of greco-latin items. This dictionary does not include the suffixes contained in the dictionary of French verbs (ELRA-L0021) and words (ELRA-L0022) such as -age, -ment, -if, -oir.
ELRA-L0027 Dictionary of French local authorities - CORA:
38,965 entries in lower cases with accents, controlled on the guide Michelin, without localities; A link can be made to the dictionary of words (ELRA-L0022) which contains inhabitants' names and their correspondence with town names.
Below is some information on the workshop 'modeling pronunciation variation for automatic speech recognition' that will be organized from 4-6 May 1998 in The Netherlands. More information about the workshop is available a http://lands.let.kun.nl/pron-var/.
advance notice ESCA Tutorial and Research Workshop on
MODELING PRONUNCIATION VARIATION FOR AUTOMATIC SPEECH RECOGNITION
4-6 May 1998
at Rolduc, a former monestary in the city of Kerkrade in the south of The Netherlands
ESCA European Speech Communication Association
COST Telecom Action 249 Continuous Speech Recognition over the Telephone
A2RT 'Automatic Acoustic Recognition Technologies' Dept. of Language & Speech Nijmegen University
TOPIC OF THE WORKSHOP
Automatic Speech Recognizers (ASR's) have improved substantially during the last decade. It has now become possible to use ASR's for many practical applications. However, when ASR's are used (and tested) under realistic conditions, the problem of pronunciation variation almost always emerges. This problem has been recognized by several research groups, and more and more effort is spen nowadays on solving this problem (see e.g. the steadily growing number of publications on this topic, especially in conference proceedings).
During this workshop we want to discuss this problem in depth and the different ways in which it could be solved. Although part of pronunciation variation is certainly language-dependent (i.e. the phonological and phonetic processes differ between languages), a large part of the variation is language independent. Furthermore, the techniques that can be used to solve this problem, i.e. to model pronunciation variation for ASR, are usually language-independent.
Up-to-date information about the workshop is available a http://lands.let.kun.nl/pron-var/.
Elizabeth Shriberg Herve Bourlard Li Deng Lori Lamel Mari Ostendorf Patti Price Roger Moore Rolf Carlson Sadaoki Furui Steve Young
Helmer Strik Dept. of Language and Speech P.O. Box 9103 6500 HD Nijmegen The Netherlands