Motivating Personality-aware Machine Translation

Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, pages 1102–1108, Lisbon, Portugal, 17-21 September 2015. c⃝2015 Association for Computational Linguistics. Motivating Personality-aware Machine Translation Shachar Mirkin∗ IBM Research - Haifa Mount Carmel, Haifa 31905, Israel shacharm@il.ibm.com Scott Nowson, Caroline Brun, Julien Perez Xerox Research Centre Europe 6 chemin de Maupertuis 38240 Meylan, France firstname.surname@xrce.xerox.com Abstract Language use is known to be influenced by personality traits as well as by socio- demographic characteristics such as age or mother tongue. As a result, it is possible to au- tomatically identify these traits of the author from her texts. It has recently been shown that knowledge of such dimensions can improve performance in NLP tasks such as topic and sentiment modeling. We posit that machine translation is another application that should be personalized. In order to motivate this, we explore whether translation preserves demo- graphic and psychometric traits. We show that, largely, both translation of the source training data into the target language, and the target test data into the source language has a detrimen- tal effect on the accuracy of predicting author traits. We argue that this supports the need for personal and personality-aware machine trans- lation models. 1 Introduction Computational personality recognition is garnering in- creasing interest with a number of recent workshops exploring the topic (Celli et al., 2014; Tkalˇciˇc et al., 2014). The addition of personality as target traits in the PAN Author Profiling challenge in 2015 (Rangel et al., 2015) is further evidence. Such user modeling – when performed on text – is built on a long-standing un- derstanding that language use is influenced by socio- demographic characteristics such as age, gender, edu- cation level or mother tongue and personality traits like agreeableness or openness (Tannen, 1990; Pennebaker et al., 2003). In this work we explore multilingual user modelling. The motivation is not only to enable modeling in multi- ple languages, but also to enable modeling multilingual users who may express different sides of their personal- ity in each language. One way to address multilingual- ity in this context is to create models separately in each language, and then fuse the resulting models. How- ever, labelled data of this nature, particularly in non- English languages, is often not available. Personality ∗ This work was mostly done while the first author was at Xerox Research Centre Europe. labelling is time consuming, requiring the completion of psychometric questionnaires which may be consid- ered invasive by many. An alternative is the use of ma- chine translation (MT) to bootstrap corpora in resource poor languages, and to translate the user’s content into a single language before modeling. Translated text, ei- ther manually or automatically generated, is known to have different characteristics than native text. Yet, MT was shown to be of use within traditional NLP tasks such as sentiment analysis (Balahur and Turchi, 2012). We explore the utility of MT for classification of demo- graphic and personality traits. MT models, even domain-specific, are user-generic. Thus, the linguistic signals …

Content truncated to 500 words