Faking a Teacher Works! Dependency Scoring Learning and Corpus Boosting for Translation-Based Cross-Lingual Dependency Parsing

Kód výsledku v IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F00216208%3A11320%2F23%3ARN9VY6ID" target="_blank" >RIV/00216208:11320/23:RN9VY6ID - isvavai.cz</a>

Výsledek na webu

<a href="https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4626681" target="_blank" >https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4626681</a>

DOI - Digital Object Identifier

<a href="http://dx.doi.org/10.2139/ssrn.4626681" target="_blank" >10.2139/ssrn.4626681</a>

Jazyk výsledku

angličtina

Název v původním jazyce

Faking a Teacher Works! Dependency Scoring Learning and Corpus Boosting for Translation-Based Cross-Lingual Dependency Parsing

Popis výsledku v původním jazyce

"Cross-lingual dependency parsing is one valuable research topic in natural language processing, where treebank translation is a prospective approach of it. Due to the divergence of languages, this method would inevitably generate noise during translation and auto-alignment. To reduce the problem, we exploit MetaNet to compute quality scores for each dependency and identify low-score ones as noise. MetaNet is a fake teacher that learns to score homework (dependencies) by comparing answers from the top student (strong parser) and the regular student (weak parser) without knowing the correct answer (gold-standard). With the scoring capability of MetaNet, we then design an iterative algorithm to boost the target treebank quality, which trains with high-quality dependencies and relabels the low-quality dependencies. We conduct experiments on Universal Dependency Treebank v2.2 to evaluate our method. Results show that our approach is highly effective with significant performance improvements across a diverse set of 10 languages. We also provide detailed analysis and discussions."

Název v anglickém jazyce

Faking a Teacher Works! Dependency Scoring Learning and Corpus Boosting for Translation-Based Cross-Lingual Dependency Parsing

Popis výsledku anglicky

"Cross-lingual dependency parsing is one valuable research topic in natural language processing, where treebank translation is a prospective approach of it. Due to the divergence of languages, this method would inevitably generate noise during translation and auto-alignment. To reduce the problem, we exploit MetaNet to compute quality scores for each dependency and identify low-score ones as noise. MetaNet is a fake teacher that learns to score homework (dependencies) by comparing answers from the top student (strong parser) and the regular student (weak parser) without knowing the correct answer (gold-standard). With the scoring capability of MetaNet, we then design an iterative algorithm to boost the target treebank quality, which trains with high-quality dependencies and relabels the low-quality dependencies. We conduct experiments on Universal Dependency Treebank v2.2 to evaluate our method. Results show that our approach is highly effective with significant performance improvements across a diverse set of 10 languages. We also provide detailed analysis and discussions."

Druh

O - Ostatní výsledky

CEP obor

—

OECD FORD obor

10201 - Computer sciences, information science, bioinformathics (hardware development to be 2.2, social aspect to be 5.8)

Projekt

—

Návaznosti

—

Rok uplatnění

2023

Kód důvěrnosti údajů

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů