Automatic Recognition of Linguistic Replacements in Text Series Generated from Keystroke Logs

Automatic recognition of linguistic replacements in text series generated from keystroke logs Daniel Couto-Vale, Stella Neumann, Paula Niemietz RWTH Aachen University, Germany K´arm´anstr. 17-19, D-52062 Aachen danielvale@icloud.com, {firstname.familyname}@ifaar.rwth-aachen.de Abstract This paper introduces a toolkit used for the purpose of detecting replacements of different grammatical and semantic structures in ongo- ing text production logged as a chronological series of computer interaction events (so-called keystroke logs). The specific case we use involves human translations where replacements can be indicative of translator behaviour that leads to specific features of translations that distinguish them from non-translated texts. The toolkit uses a novel CCG chart parser customised so as to recognise grammatical words independently of space and punctuation boundaries. On the basis of the linguistic analysis, structures in different versions of the target text are compared and classified as potential equivalents of the same source text segment by ‘equivalence judges’. In that way, replace- ments of grammatical and semantic structures can be detected. Beyond the specific task at hand the approach will also be useful for the analysis of other types of spaceless text such as Twitter hashtags and texts in agglutinative or spaceless languages like Finnish or Chinese. Keywords: translation process, keystroke logging, character chart parser 1. Introduction In a na¨ıve approach, it would be assumed that the task of translating consists in giving literal renditions of the words of the source language text in the target language, i.e. in finding equivalent words for each word in the source text and simply replacing the source text words by these target language words in the same order. However, various factors such as the incommensurability of the languages involved, differences in the situational features that determine a reg- ister, specific goals and guidelines of the translation brief among others will result in non-literal translations. Moreover, the task of mediating between two potentially different lingua-cultures also leads to translations being dif- ferent in terms of the distribution of linguistic features from original text production in the target language in several ways (Hansen-Schirra et al., 2012). Machine learning ap- proaches to translation as a sub-language (‘translationese’) report high accuracies in classifying texts as translations and non-translated texts in a mixed set based on proper- ties such as mean 1-gram length, frequencies of spellings of derivational and inflectional prefixes and suffixes, ratio between function words and content words (Volansky et al., 2015). While such studies provide evidence for the claim that translations are indeed different from non-translated texts, they do not easily support explanations for the ob- served differences. It is also assumed that some of the sta- tistical patterns are caused not by linguistic and cultural dif- ferences, but by cognitive factors such as (lack of) under- standing of the source text and fatigue by the translator. Process-oriented translation research aims at characteris- ing this cognitive activity in terms of translators’ behaviour during the translation task. For this purpose, keystroke log- gers such as Translog (Jakobsen and Schou, 1999; Carl, 2012) are used to record and replay the …

Content truncated to 500 words