Semi-supervised neural part-of-speech tagging

master's thesis

• Author: Tihana Britvić
• Mentor: Mladen Vuković (mentor)
• Committee member: Mladen Vuković (predsjednik povjerenstva)
• Committee member: Goranka Nogo (član povjerenstva)
• Committee member: Goran Igaly (član povjerenstva)
• Committee member: Luka Grubišić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Science; (Department of Mathematics); Zagreb
• Defense date and country: 2018-03-01, Croatia
• Scientific / art field, discipline and subdiscipline: NATURAL SCIENCES; Mathematics

Abstract

We present a simple method for learning part-of-speech taggers for low-resource languages using dictionaries as are reference method. Despite significant recent work, purely unsupervised techniques for part-of-speech (POS) tagging have not achieved useful accuracy required by many language processing tasks. Use of parallel text between resource-rich and resource-poor languages is one source of weak supervision that significantly improves accuracy. However, parallel text is not always available and techniques for using it require multiple complex algorithmic steps. We have shown that we can build POS-tagger by using bi-LSTMs and a freely available and naturally growing resource, the Wiktionary. Across nine languages for which we have labeled data to evaluate results, we achieve accuracy that in some cases exceeds all unsupervised methods, supervised method which uses hidden Markov chains and parallel text methods. We achieve highest accuracy reported for several languages.

Abstract (croatian)

U radu je iznesena i objašnjena jednostavna metoda za učenje morfološkog označivača za nisko resursne jezike oslanjajući se na rječnike tih jezika. Unatoč znatnom broju nedavno objavljenih radova koji oslovljavaju ovaj problem, bez nadzorne metode učenja nisu rezultirale dovoljno velikom točnošću. Jedna od metoda (slabo) nadzirnog učenja je korištenje paralelnog teksta između jezika s bogatih i siromašnim resursima koji znatno poboljšava točnost morfološkog označivanja. Međutim, paralelni tekstovi nisu uvijek dostupni, a tehnike za upotrebu istog zahtijevaju mnogo složenih algoritamskih koraka. U radu smo pokazali kako izgraditi jednostavan morfološki označivač pomoću bi-LSTM neuronskih mreža i slobodno dostupnog i prirodno rastućeg resursa, Wiktionary-a Za devet jezika koje smo označili podatke u svrhu procjene dobivenih rezultata, postižemo točnost koja u nekim slučajevima prelazi sve metode bez nadzora i metode s nadzorom koje koriste skrivene markovljeve lance i paralene korpuse.

• Keywords: part-of-speech taggers; low-resource language; bi-LSTM; neural network; Witkionary
• Keywords (croatian): morfološki označivač; nisko resursni jezik; bi-LSTM; neuronske mreže; Wiktionary
• Language: english
• URN:NBN: urn:nbn:hr:217:169195
• Study programme: Title: Computer Science and Mathematics; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra računarstva i matematike (magistar/magistra računarstva i matematike)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Open access
• Created on: 2018-07-11 09:49:32