Predicting default and non-default aspectual coding: Impact and density of information features

Michael Richter, Tariq Yousef

Research output: Contribution to journalConference articleResearchpeer-review

Abstract

This paper presents a study on the automatic classification of default and nondefault codings for aspect-marked verbs in six Slavic and one Baltic language. As classifier a Support Vector Machine (SVM) and as verbal features Shannon Information (SI) and Average Information Content (IC) have been utilised. In all languages high accuracy of the classification has been achieved. In addition, we found indications for the validity of the Uniform Information Density principle within SI and IC.

Original languageEnglish
JournalCEUR Workshop Proceedings
Volume2521
ISSN1613-0073
Publication statusPublished - 2019
Event3rd Workshop on Natural Language for Artificial Intelligence, NL4AI 2019 - Rende, Italy
Duration: 19. Nov 201922. Nov 2019

Conference

Conference3rd Workshop on Natural Language for Artificial Intelligence, NL4AI 2019
Country/TerritoryItaly
CityRende
Period19/11/201922/11/2019

Bibliographical note

Funding Information:
* Funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundatino) project number: 357550571.

Publisher Copyright:
Copyright © 2019 for this paper by its authors.

Keywords

  • Coding
  • Information content
  • Verb aspect

Fingerprint

Dive into the research topics of 'Predicting default and non-default aspectual coding: Impact and density of information features'. Together they form a unique fingerprint.

Cite this