We will be presenting the results of our first study at the UCCTS 2020 (Using Corpora in Contrastive and Translation Studies) conference in Bertinoro (Italy) on 7-9 September 2020 (if the conference takes place in the current coronavirus reality). Here’s the abstract of our presentation:
Cross-linguistic similarities in lexis – examining cognate activation through temporal and accuracy data from the Polish Interpreting Corpus (PINC)
The objective of this corpus-based study is to gain novel insight into the mechanism of lexical activation that underlie bilingual language control in conference interpreting on the basis of naturalistic data. We wanted to examine if and to what extent professional interpreters take advantage of cross-linguistic similarities in lexis (whether they interpret cognates by applying their cognate or non-cognate equivalents).
Activation is one of the central mechanisms of bilingual control, largely examined on populations of bilinguals and manifested frequently through the cognate facilitation effect (Dijkstra, Van Hell, & Brenders, 2014). Such findings are easily explained by the Bilingual Interactive Activation Plus (BIA+) model (Dijkstra & van Heuven, 2002), currently extended into the Multilink model (Dijkstra et al., 2018), which posits language non-selective lexical access and a multiplicity of cross-level activations. As a result, seeing the English word reduction will activate the Polish cognate redukcja due to orthographic and phonological similarity. These words in turn will activate their shared semantic representation and the word redukcja will have a greater activation than zmniejszenie, the Polish word with a similar meaning but a different form.
The mechanism of lexical activation is largely in line with the theory of translation that suggests two strategies used in translation and interpreting (de Groot & Christoffels, 2006; Macizo & Bajo, 2006; Ruiz, Paredes, Macizo, & Bajo, 2008). One is vertical or conceptually mediated translation, according to which the source language message is first decoded and its conceptual representation is activated through phonological, morphological and semantic analysis and then its lexical representation in the target language is activated for production. The other is horizontal or structurally mediated translation in which source language utterances are directly transcoded into their target language equivalents thanks to memory associations (de Groot & Christoffels, 2006). This account is also in line with the literal translation hypothesis (Halverson, 2015) and the recursive model of translation (Schaeffer & Carl, 2013). Interpreters might use the horizontal or vertical strategy depending on context. We predict that if they use cognate rather than non-cognate equivalents when interpreting, especially in combination with a short ear-voice span (EVS), they opt for the horizontal strategy and use the most activated target language equivalent in accordance with the Multilink model. Alternatively, if they use non-cognate equivalents for cognates they favour the vertical strategy and use more inhibition to suppress the activated cognate equivalent. Defrancq (2015) found that cognates interpreted by means of their cognate equivalents in the target language triggered a shorter EVS than other words elsewhere in the corpus. Hansen-Schirra, Nitzke, and Oster (2017) determined that the use of cognate translation equivalents in the written translation corpus was modulated by numerous factors, such as context, language status and translation experience. Our study extends this research by examining to what extent interpretation of cognates is modulated by word frequency and the number of translation equivalents and how EVS changes depending on the strategy used.
In line with Defrancq (2015), we predicted that cognates would generate a shorter ear-voice span (EVS) than control words since the former would enjoy facilitation. We also predicted shorter EVS for cognate than non-cognate translations in line with the horizontal translation account. We expected modulation of the EVS by such factors as frequency and number of translation equivalents: we predicted shorter EVS for more frequent items and longer EVS for items with more translation equivalents for non-cognate translations and no effect for cognate translations.
In an attempt to test our predictions on naturalistic data, we created PINC (Polish Interpreting Corpus) – a new bidirectional, parallel and comparable time-annotated corpus of interpretations performed by professional conference interpreters in the European Parliament. It includes over 100,000 words and is divided into four subcorpora: Polish original speeches, their English interpretations, English original speeches, their Polish interpretations. We identified cognate nouns represented in the subcorpus of Polish original speeches and aligned them with data on frequency (Mandera, Keuleers, Wodniecka, & Brysbaert, 2015), number of translation equivalents, source text speed, interpretation speed and location in the source text. We then examined their translations in the parallel subcorpus of their interpretations into English. We evaluated translation accuracy and calculated ear-voice spans, or time lags between these words in the source and target texts.
Preliminary data shows the expected cognate facilitation effect in translation accuracy data only and not in temporal data. In other words, cognates were interpreted more accurately than controls but there was no difference in EVS. Also, cognates were predominantly interpreted by means of their cognate equivalents, but again there was no difference in EVS generated by cognate and non-cognate equivalents. EVS was not modulated by frequency, but it was affected by the number of translation equivalents as words with more translation equivalents generated longer EVS. Our temporal data results are at a variance with Defrancq (2015) and do not support the Multilink model, which might be due to the different interpreting direction in that study and ours. However, our accuracy data do support the horizontal translation account. The lack of support for cognate facilitation in the EVS data might have various explanations. First, sentence context and language proficiency have been found to decrease cognate facilitation (Bultena, Dijkstra, & van Hell, 2014; Dijkstra et al., 2014; Libben & Titone, 2009) while our participants were highly proficient in their working languages and processed cognates in sentence and text contexts. Also, the psycholinguistic models of cognate processing are largely based on visual presentation of stimuli, while our data comes predominantly from auditory presentation. Finally, we obtained few datapoints for non-cognate translations of cognates, which may have been insufficient to find the effect.
The study employs a corpus-based paradigm and shows support for the horizontal strategy to interpret cognates in product data (accuracy) but not in process data (EVS). It thus contributes to our knowledge on how language control issues are managed by experienced interpreters in extreme conditions of high cognitive load and temporal constraints.
Bultena, S., Dijkstra, T., & van Hell, J. G. (2014). Cognate effects in sentence context depend on word class, L2 proficiency, and task. The Quarterly Journal of Experimental Psychology, 67(6), 1214-1241.
de Groot, A. M. B., & Christoffels, I. K. (2006). Language control in bilinguals: Monolingual tasks and simultaneous interpreting. Bilingualism-Language and Cognition, 9(2), 189-201. doi:10.1017/s1366728906002537
Defrancq, B. (2015). Corpus-based research into the presumed effects of short EVS. Interpreting, 17(1), 26-45. doi:10.1075/intp.17.1.02def
Dijkstra, T., Van Hell, J. G., & Brenders, P. (2014). Sentence context effects in bilingual word recognition: Cognate status, sentence language, and semantic constraint. Bilingualism: Language and Cognition, 1-17. doi:10.1017/s1366728914000388
Dijkstra, T., & van Heuven, W. J. B. (2002). Modeling bilingual word recognition: Past, present and future. Bilingualism: Language and Cognition, 5(03). doi:10.1017/s1366728902283017
Dijkstra, T., Wahl, A., Buytenhuijs, F., Van Halem, N., Al-Jibouri, Z., De Korte, M., & RekkÉ, S. (2018). Multilink: a computational model for bilingual word recognition and word translation. Bilingualism: Language and Cognition, 1-23. doi:10.1017/s1366728918000287
Halverson, S. (2015). Cognitive translation studies and the merging of empirical paradigms: The case of ‘literal translation’. Translation Spaces, 4(2), 310-340.
Hansen-Schirra, S., Nitzke, J., & Oster, K. (2017). Predicting cognate translation. In S. Hansen-Schirra, O. Czulo, & S. Hofmann (Eds.), Empirical modelling of translation and interpreting(pp. 3-22). Berlin: Language Science Press.
Libben, M. R., & Titone, D. A. (2009). Bilingual lexical access in context: evidence from eye movements during reading. J Exp Psychol Learn Mem Cogn, 35(2), 381-390. doi:10.1037/a0014875
Macizo, P., & Bajo, M. T. (2006). Reading for repetition and reading for translation: do they involve the same processes? Cognition, 99(1), 1-34. doi:10.1016/j.cognition.2004.09.012
Mandera, P., Keuleers, E., Wodniecka, Z., & Brysbaert, M. (2015). Subtlex-pl: subtitle-based word frequency estimates for Polish. Behav Res Methods, 47(2), 471-483.
Ruiz, C., Paredes, N., Macizo, P., & Bajo, M. T. (2008). Activation of lexical and syntactic target language properties in translation. Acta Psychol (Amst), 128(3), 490-500. doi:10.1016/j.actpsy.2007.08.004
Schaeffer, M., & Carl, M. (2013). Shared representations and the translation process: A recursive model. Translation and Interpreting Studies, 8(2), 169-190. doi:10.1075/tis.8.2.03sch