人工智能论文英文版-Step-by-step Instructions and a Simple Tabular Output Format Improve the Dependency Parsing Accuracy of LLMs.pdfVIP

Step-by-stepInstructionsandaSimpleTabularOutputFormatImprove

theDependencyParsingAccuracyofLLMs

HiroshiMatsudaChunpengMaMasayukiAsahara

MegagonLabs,Tokyo,NationalInstituteforJapanese

RecruitCo.,Ltd.LanguageandLinguistics

{hiroshi_matsuda,ma.chunpeng}@megagon.aimasayu-a@ninjal.ac.jp

Abstract

5Recentadvancesinlargelanguagemodels

2(LLMs)haveenabledimpressiveperformance

0invarioustasks.However,standardprompt-

2ingoftenstrugglestoproducestructurallyvalid

nandaccurateoutputs,especiallyindependency

uparsing.Weproposeanovelstep-by-step

Jinstructionstrategy,whereuniversalpart-of-

1speechtaggingprecedesthepredictionofsyn-

1tacticheadsanddependencylabels,anda

]simplifiedCoNLL-Ulikeoutputformat,our

Lmethodachievesstate-of-the-artaccuracyon

CUniversalDependenciesdatasetsacross17lan-

s.guageswithouthallucinationorcontamination.

cWefurthershowthatmultilingualfine-tuning

[simultaneouslyimprovescross-languagegener-

1alizationperformance.Ourresultshighlight

vtheeffectivenessofexplicitreasoningsteps

3inLLM-basedparsingandofferascalable,

8format-consistentalternativetobracket-based

9approaches.Figure1:Frameworkoft