"Don't worry, it's just noise": quantifying the impact of files treated as single textual units when they are really collections - Archive ouverte HAL Access content directly
Conference Papers Year : 2021

"Don't worry, it's just noise": quantifying the impact of files treated as single textual units when they are really collections

(1, 2, 3, 4)
1
2
3
4

Abstract

Literature works may present many autonomous or semi-autonomous units, such as poems for the first or chapter for the second. We make the hypothesis that such cuts in the text's flow, if not taken care of in the way we process text, have an impact on the application of the distributional hypothesis. We test this hypothesis with a large 20M tokens corpus of Latin works, by using text files as a single unit or multiple "autonomous" units for the analysis of selected words. For groups of rare words and words specific to heavily segmented works, the results show that their semantic space is mostly different between both versions of the corpus. For the 1000 most frequent words of the corpus, variations are important as soon as the window for defining neighborhood is larger or equal to 10 words.
Fichier principal
Vignette du fichier
dont_worry_just_noise_ACL.pdf (1.23 Mo) Télécharger le fichier
Origin : Files produced by the author(s)

Dates and versions

hal-03481620 , version 1 (15-12-2021)

Identifiers

  • HAL Id : hal-03481620 , version 1

Cite

Thibault Clérice. "Don't worry, it's just noise": quantifying the impact of files treated as single textual units when they are really collections. Workshop on Natural Language Processing for Digital Humanities (NLP4DH), NLP Association of India (NLPAI), Dec 2021, NIT Silchar, India. pp.95-105. ⟨hal-03481620⟩
96 View
37 Download

Share

Gmail Facebook Twitter LinkedIn More