Machine-translated texts as an alternative to translated dictionaries for LIWC

Abstract

Linguistic Inquiry and Word Count (LIWC) is a text analysis program developed by James Pennebaker and colleagues. At the basis of LIWC is a dictionary that assigns words to categories. This dictionary is specific to English. Researchers who want to use LIWC on non-English texts have typically relied on translations of the dictionary into the language of the texts. Dictionary translation, however, is a labour-intensive procedure. In this paper, we investigate an alternative approach: to use Machine Translation (MT) to translate the texts that must be analysed into English, and then use the English dictionary to analyse the texts. We test several LIWC versions, languages and MT engines, and consistently find the machine-translated text approach performs better than the translated-dictionary approach. We argue that for languages for which effective MT technology is available, there is no need to create new LIWC dictionary translations.

Performing Multilingual Analysis With Linguistic Inquiry and Word Count 2015 (LIWC2015). An Equivalence Study of Four Languages

“…However, studies such as the one conducted by Araújo et al ( 2020 ) on 14 tools have started to show promising results that such a strategy could work. Likewise, Boot ( 2021 ), using several LIWC versions, multiple machine translation engines, and replicating the results in different languages (i.e., Dutch, German, and Spanish), suggested that the machine-translation approach might be better than translating the dictionary, at least for some languages. Thus, we argue that the validity of using an automatic method for translating the input text and using the translation for language analysis is an exciting topic for future studies about multilingual analysis.…”

Development of the Japanese Version of the Linguistic Inquiry and Word Count Dictionary 2015

“…Nonetheless, this approach may miss high-frequency words observed only in source languages. At least for the present moment, the coverage of this approach is limited to texts that do not include informal words or culture-specific expressions (see also Boot, 2021 , for further discussion comparing machine-translated and dictionary-based approaches).…”

Beyond English: Considering Language and Culture in Psychological Text Analysis

“…If we focus on cross-language adaptation of closed-vocabulary methods, it should be emphasized that these tools are naturally based on the specifics of the source (original) language for which they were developed, most often English [see Mehl (2006)]. Therefore, adapting such dictionaries to other languages is often a complicated and time-consuming process that faces a series of additional challenges (Bjekić et al, 2014;Dudãu and Sava, 2020;Boot, 2021). First, the methods are most often based on the original cultural and linguistic structure rather than the target culture or language, that is on the imposed-etic approach (Berry et al, 1997).…”

“…Several strategies have been used to adapt the LIWC dictionary to other languages (Boot, 2021). These include the supervised translation of the English dictionary word by word (Bjekić et al, 2014;Dudãu and Sava, 2020), the use of the existing word corpora and their assignment to corresponding LIWC categories (Andrei, 2014) or as an enrichment of LIWC categories (Gao et al, 2013;Meier et al, 2019), the use of dictionaries in closely related languages (Massó et al, 2013), the modification of the older version of the dictionary (Zijlstra et al, 2004), or adapting the original dictionary via machine translation (Van Wissen and Boot, 2017).…”

“…The translation and adaptation process faces most of the issues described above. Here, however, the analysis deals also with additional challenges, connected to level of morphology and syntax of the target languages, for example the pronoundrop phenomenon (in some languages, users very frequently omit pronouns, particularly in their subject positions; e.g., "tengo hambre" in Spanish dropping the first-person singular pronoun "yo") ( Świ ątek, 2012), grammatical classification (e.g., pronominal adverbs in Dutch, that combine pronouns/adverbs with prepositions-"we doken erin" which replaces "we doken in het"-"we dived into it"), grammatical restrictions (some linguistic features are restricted to particular languages, see below), with case sensitivity problems (LIWC is not case-sensitive which makes it difficult to process certain words, e.g., the German word "Sie" which, if capitalized, serves as formal second person singular or plural pronoun and, when not capitalized, serves as third person plural pronoun), and the above mentioned ambiguity (including, if the capitalized word appears at the beginning of a sentence) (Boot, 2021).…”