Automatically stress labelled morphological lexicon Sloleks 1.2, version 1.1

  Avtorji

Krsnik, Luka ; Robnik-Šikonja, Marko ; Šef, Tomaž and Krek, Simon

  Identifikator vnosa

http://hdl.handle.net/11356/1186

 URL projekta

https://gitea.cjvt.si/lkrsnik/stress_asignment

 Dokumentirano v

http://videolectures.net/jota_krsnik_napovedovanje_naglasa/
https://repozitorij.uni-lj.si/IzpisGradiva.php?id=98276

 Datum objave

2018-05-08

 Vrsta

lexicalConceptualResource, text

 Velikost

2774745 words, 100805 entries

 Jezik(i)

Slovenian

 Opis

This lexicon is an extended version of Sloleks 1.2, http://hdl.handle.net/11356/1039. It contains all the original data from Sloleks with added information about the stress of each word form, which is included in two ways: information about stress location only, and information about stress location and type. Stress assignment was performed automatically, with algorithms based on deep neural networks which correctly predicted accent location in 91.5% and combined accent type and location in 88.5% of test data. Therefore not all accents are correct. This updated 1.1 version of the lexicon contains stress asignments with an improved algorithm, which reduces the error by about 1% against the previous 1.0 version.

 Izdajatelj

Faculty of Computer and Information Science, University of Ljubljana

 

Centre for Language Resources and Technologies, University of Ljubljana

 Ključne besede

word stress

 Zbirke

CLARIN.SI data & tools

 Druge različice

Seznam vseh različic  

Prikaži polni zapis vnosa