SADiLaR Language Resource Repository
Not a member yet
536 research outputs found
Sort by
Lwazi II Proper Name Call Routing Telephone Corpus
No full textShort prompts of proper names and language names collected via the telephone network
CKarma
No full textCKarma is a compound analyser for Afrikaans, to be used for the detection of word boundaries within compounds. It takes as input a string, and produces as output an analysed string, without any tags. For example, the string "hondehokdak" ('dog house roof') will be analysed as "hond _ e + hok + dak", where the plus sign indicates the beginning of an independent constituent, and the underscore the beginning of a dependent constituent (i.e. a valence morpheme). CKarma is a C5 classifier, trained on data consisting of circa 47,000 compound and 7,000 non-compounds. The resulting decision tree and cases can be converted to C code by means of a script written by MM van Zaanen. This C code can then be implemented in any other system
South African English Language Pack
No full textSpeech recognition language pack (language models grammars, lexicons, ) for use with TeliSpeech ™ of Telism
PSearch 1.1.
No full textPSearch is based on Paramsearch, a tool created by Antal van den Bosch for automatic algorithmic parameter optimisation for TiMBL and other machine learning algorithms. Paramsearch is available for download from http://ilk.uvt.nl/software/. PSearch works basically on the same principles as Paramsearch, with the major difference being the way that the sizes of the progressive training and evaluation data sets are generated. Another difference is that PSearch can utilise all of the TiMBL algorithms
AuCoPro Splitting Dataset
No full textThe AuCoPro Splitting dataset contains compounds annotated with their compound boundaries and linking morphemes for Afrikaans and Dutch
Ragel
No full textRagel was developed by using traditional methods for stemming/lemmatisation (i.e. affix stripping), and consists of language-specific rules for identifying word-forms in the lexicon of the spelling checker. However, Ragel cannot be considered either a "pure" lemmatiser or a "pure" stemmer in the true sense of the word, since it was developed specifically for purposes of spelling checking. In this sense, both derived and inflected word-forms that are not in the lexicon of the spelling checker are analysed by Ragel, only until a word in the lexicon is found, whether that word is a lemma or not (e.g. "ontbossing" 'deforestation' will be analysed as "ontbos" 'deforest' and not necessarily as "bos" 'forest'). Moreover, Ragel does not cover the full complexity of Afrikaans morphology, but only a subset that was considered (a) highly productive; and (b) unproblematic in terms of morphonological changes. Ragel has been implemented in Afrikaanse Speltoetser 3.0, Tesourus 1.0 en Woordafbreker
Pretoria Sepedi Corpus (Gold Standard)
No full textA section of the Pretoria Sepedi Corpus for POS, manually checked for POS tags
Tshivenda Speech Corpus
No full textA telephone speech database for Tshivenda ASR system was created from transcriptions of speech data collected. Prompted and some some speec