Publication type: Conference paper
Type of review: Not specified
Title: Assessing keyness using permutation tests
Authors: Mildenberger, Thoralf
et. al: No
Conference details: Statistical standards for scientific discovery in linguistics, Zurich, 4–6 October 2017
Issue Date: 6-Oct-2017
Language: English
Subjects: Corpus statistics; Keyword analysis; Permutation test
Subject (DDC): 400: Language, linguistics
500: Natural sciences and mathematics
Abstract: In corpus linguistics, statistical hypotheses tests (e.g. Likelihood-Ratio, Chi-Square or Fisher’s Exact Test) are used for identifying keywords, i. e. words that occur more frequently in one corpus than in another one. A problem with these tests is that they are all essentially based on the same, often inappropriate sampling model: Corpora are modeled as sets of independently sampled tokens, although in many cases the natural sampling units are whole texts. Occurrences of words tend to cluster in only a few texts, and in the extreme a word may be identified as a keyword because it appears very often in only one single text. We propose the use of permutation tests based on a model that regards corpora as samples of texts instead of samples of tokens, which often seems much more realistic. P-values for assessing keyness can be obtained using Monte Carlo methods, making the method applicable in practice. We outline our approach and contrast the results with those obtained by traditional methods.
Fulltext version: Published version
License (according to publishing contract): Licence according to publishing contract
Departement: School of Engineering
Organisational Unit: Institute of Data Analysis and Process Design (IDP)
Published as part of the ZHAW project: Energiediskurs Messen
Appears in Collections:Publikationen School of Engineering

Files in This Item:
There are no files associated with this item.

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.