UTILISING SOCIAL MEDIA THROUGH CROWDSOURCING FOR MORPHOLOGICAL RESOURCES ACQUISITION OF UNDER-RESOURCED LANGUAGE (U-RL): MEl.ANAU

VOON, MEl WEl (2015) UTILISING SOCIAL MEDIA THROUGH CROWDSOURCING FOR MORPHOLOGICAL RESOURCES ACQUISITION OF UNDER-RESOURCED LANGUAGE (U-RL): MEl.ANAU. [Final Year Project Report] (Unpublished)

	PDF Voon Mei Wei 24pgs.pdf Download (5MB)
	PDF (Please get the password from TECHNICAL & DIGITIZATION MANAGEMENT UNIT, ext: 082-583913/ 082-583914) Voon Mei Wei latestft.pdf Restricted to Registered users only Download (28MB)

Abstract

Morphological analyser is the finl processing 1001 required in Nalural Language Processing To anct/yse structure ofa word, the analyser needs 1I100phologica/ resources The resourCeS are from dictionary. grammar book6), and wrilft:n lexls. Yei. huw fa acquire morphological rl!sources jor under-res()urced languages knOllling fha! the /cmglloges are criljca/ly locking of malerials:1 in curren! approach. morphological reSOurCeS an: ucquirf:1d (rom hardcopy versions wherttby (lilt! needs (0 digiNse Ihe documents infO sofh:opy versinns. Dlie to d~[ficlllly in digilisa/iun us if IS lim~ consuming and expensive, 'his project is propO~'ing 11 workflow of acql/iring morphvlogicol resources jor lInder-resourced languages, in the case of Jvle/UI1CIlI language, hy ulili!.·ing social media. Three main stages in the work are: i) c:rowdsourcing the social media hy using CI weh crawler Spider 1.1 and Jsoup methud: ii) performing hybrid normalisalion to (rcm~furm the crawled dolO with informollll1d noisy nature il1lo a cle(med wordlisl: ;il) validating the wordlist, is a crucial stage due fo languages mixing that causes uncertainty of spelling standard. AI thiS sfCl¢e, edIt distance similarilyalgorithms, Jaro-Winkler distance, Levenshtein-based distance, and N-grwn distam.'c, are applied to ;den/~fy the spelling !.'/andord be/ween a source word from Ihe wordlis/ and a largel word in the dictionary. The resulls shOl.." fhal Jaro-Winkler pet/orms the best compared /0 the other two algorithms becc/llse il returns Ihe highes! F-score and the longesl validaled It'ordlisl. The l'ulidaled wordli.\'ts are then considered ns {he A1elol1ou morp/wtogical reSOJ/J"Ci?'i !hal Clln he apfJ"ed I~v computational linguists in the computCllional morphology. indirectly, Ihl! proposed workflow can also bi; usal (0 acquire morphological ri;SOLlrtes for other 1./J1c!i;r-resourced languages in Sura wok

Actions (For repository members only: login required)

View Item