Bootstrapping a multilingual part-of-speech tagger in one person-day

Abstract

This paper presents a method for bootstrapping a fine-grained, broad-coverage part-of-speech (POS) tagger in a new language using only one personday of data acquisition effort. It requires only three resources, which are currently readily available in 60-100 world languages: (1) an online or hardcopy pocket-sized bilingual dictionary, (2) a basic library reference grammar, and (3) access to an existing monolingual text corpus in the language. The algorithm begins by inducing initial lexical POS distributions from English translations in a bilingual dictionary without POS tags. It handles irregular, regular and semi-regular morphology through a robust generative model using weighted Levenshtein alignments. Unsupervised induction of grammatical gender is performed via global modeling of contextwindow feature agreement. Using a combination of these and other evidence sources, interactive training of context and lexical prior models are accomplished for fine-grained POS tag spaces. Experiments show high accuracy, fine-grained tag resolution with minimal new human effort.

Bootstrapping POS taggers using unlabelled data

“…There has been some previous work on boostrapping POS taggers (e.g., Zavrel and Daelemans (2000) and Cucerzan and Yarowsky (2002)), but to our knowledge no previous work on co-training POS taggers.…”

Arabic Tweets Treebanking and Parsing: A Bootstrapping Approach

“…Bootstrapping is used to create labelled training data from large amounts of unlabelled data (Cucerzan and Yarowsky, 2002).…”

Resource‐Light Approaches to Computational Morphology Part 1: Monolingual Approaches

“…2008; Oflazer et al. 2001), and manual encoding of basic linguistic facts (e.g., Cucerzan and Yarowsky 2002; Feldman and Hana 2010; Tepper and Xia 2010). Learning from a different language (e.g., Bosch et al.…”

“…Learning from a different language (e.g., Bosch et al. 2008; Cucerzan and Yarowsky 2002; Feldman and Hana 2010), another resource‐light strategy, will be discussed in our forthcoming survey (Feldman and Hana forthcoming).…”