Machine Translation for EFL

Paul J. Moore (University of Queensland), Phil Murphy, Luann Pascucci, and Scott Sustenance (Kanda University of International Studies)

Moore, P.J., Murphy, P., Pascucci, L., & Sustenance, S. (2019). Machine translation for EFL. Relay Journal, 2(1), 228-235. https://doi.org/10.37237/relay/020127

[Download paginated PDF version]

*This page reflects the original version of this document. Please see PDF for most recent and updated version.

Abstract

This paper reports on an ongoing study into the affordances of free online machine translation for students learning English as a foreign language (EFL) at the tertiary level in Japan. The researchers are currently collecting data from a questionnaire, task performance, and interviews with 10-15 EFL learners in an English Language Institute in a university in Japan. The paper provides some background on the changing role of translation in language learning theory and pedagogy, before focusing literature related to technical developments in machine translation technology, and its application to foreign language learning. An overview of the research methodology is provided, along with some insights into potential findings. Findings will be presented in subsequent publications.

Keywords: online machine translation, EFL, Japanese university

 

The study reported here investigates the interaction between language learner, technology and task in evaluating the affordances of Google Translate in second language task performance (Chapelle, 2001), to ascertain how the technology mediates task performance in a Japanese university context, across tasks of varying complexity and modality. In line with the broader focus on how second language production and interaction are mediated by technology in the field of computer-assisted language learning there have been recent calls to investigate how computer-based speaking performance is influenced by learners’ interaction through and with technology (Chung, 2017; Iwashita, May & Moore, 2017). Reflecting on how communication is increasingly technologically mediated in broader society (and second language learning), Chapelle and her colleagues (e.g., Sauro & Chapelle, 2017) argue that the construct of L2 communicative competence should be broadened to include technological competence in cultural contexts. By investigating the complex interactions among test-taker, task and technology in a Japanese context, this study has the potential to provide further insights into technologically mediated communicative competence.

Literature Review

The role of translation in foreign language learning

Informed by the prominence of communicative language teaching (CLT) in second language acquisition (SLA) theory, as well as a general rejection of traditional grammar translation methodology (e.g., Howatt, 1984), translation was generally discounted as a pedagogical tool from a cognitive SLA perspective. This perspective has included an assumption that maximal exposure to the second language is necessary for second language development, and that use of the first language in second language pedagogy interferes with the development of second language proficiency. Recent perspectives on second language learning (e.g., Cook, 2010, Moore, 2013, 2017, 2018; Ortega, 2014), have led to a reconsideration of the supporting roles that the L1 play in the development of second (bi/multilingual) language proficiency.

Online machine translation

As with the introduction of all new technologies, there is a long held distinction between the hype, often created in the first instance by the creators of the technology (including MT language service providers), and the reality. Historically, the hype of emerging technologies has been met with polarizing views on their potential influence on society, and education. From the field of translation education, Cronin (2013, p. 2) distinguishes between ‘the backwards look’, which laments the lost skills (and jobs) brought about by automation and ‘the forwards look’ which sees technological advances as miraculous panacea. In tracing the development of machine translation, from IBM’s 1954 ‘electronic brain’ Georgetown experiment to Google’s 2016 introduction of neural machine translation (NMT), Moorkens (2018) argues that that there has been some improvement in accuracy, though the hype surrounding NMT is somewhat responsible for a boom in its by translation service providers. In professional translation, MT is commonly used in conjunction with human translation, and May (2018) draws on the concept of ‘fitness for purpose’ to establish conditions regarding when and how MT may be utilised. This concept is also useful for second language learning and use, in that critical awareness of the affordances of such tools may be important in determining the conditions under which they may support language learning.

Online machine translation and foreign language learning

Little research has been undertaken into the use of online machine translation (OMT) for foreign language learning, and the existing research generally relies on self-report data. Such an approach suggests the exploratory nature of this relatively new research area. Recent research has focused on the role of MT in post-editing of academic writing (O’Brien et al., 2018) for EFL writers, finding that participants had mixed views as to the effectiveness of MT in comparison with self-editing of their written work. Groves & Mundt (2015) argue that regardless of teachers’ and researchers’ perspectives, learners of English will draw on tools like Google Translate if they perceive them as useful. Briggs (2018) reports on a survey of 80 Korean learners of EFL regarding their use of and perceptions as to the effectiveness of OMT in supporting their language learning. Approximately half of the participants reported believing that OMT was beneficial for language learning, and that their use should be permitted in EFL classrooms.

The fact that learners perceive OMT favourably supports their continued use for language learning purposes, and learners and teachers should be critically aware of the affordances of such tools. Interestingly, learners in recent studies reported above seem to be less aware of the capability of OMT to analyse spoken language, and its speech-to-text capabilities, the former of which was a major focus of the hype generated during the launch of Google’s Pixel 2 in October 2017.

Research Questions

RQ1. How does GT mediate learners’ performance on a range of second language tasks, varying in complexity and modality?

1. A one-way oral task (giving directions);
2. A multimodal task (interpreting a sign);
3. A writing task (based on prompts);
4. A narrative task (recounting experience).

RQ2. How effectively does GT perform in translating language learners’ L1 or L2 output in each task?

RQ3. How effective is GT in supporting (a) task performance and (b) independent language learning, as perceived by the learners.

Overview of the Methodology

The research follows a qualitative design with a focus on an empirical evaluation of the affordances of Google Translate (GT) for independent CALL (Chapelle, 2001; Levy & Moore, 2018). Data from four sources will be triangulated to answer the three research questions. In addition to allowing for data collection over a short period of time, such a design also allows for validation of findings from a range of perspectives.

Participants

The study is being conducted in an English Language Institute in a small private university in Japan, where students major in languages as well as related programs in international studies. Data are being collected from ten first or second year students of English as a foreign language (aged 18-21), enrolled in English language courses at the university. As such, recruitment will be based on a convenience sample of potential participants who fit the purpose of the study (Dörnyei & Taguchi, 2010). Student participants’ language proficiency will be at the intermediate level (A2 or B1 on the internationally recognised Common European Framework for Languages CEFR).

Data collection

This study will draw upon three major data sources:

• A background survey
• Video-recorded reflection meetings between students and learning advisors (both LAs are co-investigators on the project);
• A follow-up semi-structured interview, where participants reflect on their experience with the technology

Data analysis

Background survey. Survey data will provide background data on participants’ age, language learning experience, second language proficiency, and experience with and preliminary perceptions of OMT technologies.

Analysis of recorded interactions. (i) GT output: Following Briggs (2018) and Moorkens et al. (2018) the quality of the GT output will be analysed for quality, including errors in inflection morphology, word order, omission, addition, or mistranslation. (ii) Learners’ task-based technology-mediated interaction: The recorded interactions will be analysed inductively for instances where and how the learners use the technology, and how the technology may influence the interaction. This analysis will be triangulated against learners’ perceptions provided in the follow-up interviews.

Analysis of interview data. Interview data will be coded and analysed collaboratively in an iterative process, using NVivo10. Transcripts will be read and re-read independently by two of the researchers before meeting to compare, reach agreement, and develop categories which represent both expected themes and those arising from the data. The approach taken here, following King and Horrocks (2010), is to code for concepts arising from the framework provided by the research questions (and the literature they were based on), and for those which emerge from the data (to be initially tracked through the use of memos in NVivo), and then to cluster related codes into themes representing aspects of the data more broadly (Miles & Huberman, 1994).

Preliminary Observations

Data analysis will commence after the completion of data collection in January, 2019. It is expected that all learners will be aware of the text translation capability of GT, but they will be less familiar with the voice recognition and augmented-reality-supported camera functions. Text-based translation is expected to be more accurate than other functions, which convert audio or visual data to text before translation occurs. Across all modalities, it is expected that increases in task and language complexity will result in reduced accuracy. It is also expected that learners will trust the accuracy of the output of GT, except for glaring errors, in line with their levels of language proficiency. Learners’ feedback on the affordances and appropriacy of using GT in their language learning are expected to be influenced by their prior experience with similar tools, as well as personal and institutional language ideologies and policies.

Notes on the Contributors

Paul J. Moore lectures in and coordinates the Master of Applied Linguistics program in The School of Languages and Cultures, University of Queensland. His research interests include task-based interaction in face-to-face and technologically mediated contexts, intercultural discourse, and the dynamic roles of the L1 in L2 interaction. His research interests include sociocognitive perspectives on task-based interaction in classroom and online contexts, intercultural communication, and the dynamic roles of the L1 in L2 interaction. P.moore5@uq.edu.au

Phil Murphy is Professor of English and Director of the English Language Institute at Kanda University of International Studies. His research interests include computer-mediated interaction and feedback, collaborative reading, as well as the application of augmented reality to second language learning.

Luann Pascucci is a Senior Lecturer at Kanda University of International Studies. She has a Master of Arts in Linguistics from the University of South Florida, and her research interests include digital game-based language learning, CLIL/CBLT, translation, and pragmatics.

Scott Sustenance is a Senior Lecturer in the English Language Institute at Kanda University of International Studies in Japan. He has a Master of Arts in Applied Linguistics from Griffith University, Australia, and his research interests include vocabulary retention, linguistic landscapes and technology in the classroom.

References

Briggs, M. (2018). Neural machine translation tools in the language learning classroom: Students’ use, perceptions, and analyses. JALT CALL Journal 14(1), 3-24.

Chapelle, C. A. (2001). Computer applications in second language acquisition: Foundations for teaching, testing and research. Cambridge: Cambridge University Press.

Chung, Y.-R. (2017). Validation of technology-assisted language tests. In C. Chapelle & S Sauro (Eds.), The Handbook of technology and second language teaching and learning (pp. 332-347). Hoboken, NJ: John Wiley & Sons, Inc.

Cook, G. (2010). Translation in language teaching. Oxford: Oxford University Press.

Cronin, M. (2013). Translation in the digital age. London and New York: Routledge.

Dörnyei, Z., & Taguchi. T. (2010). Questionnaires in second language research (2^nd ed.). London and New York: Routledge.

Groves, M., & Mundt, K. (2015). Friend or foe? Google Translate in language for academic purposes. English for Specific Purposes, 37, 112-121. doi: 10.1016/j.esp.2014.09.001

Howatt, A. (1984). A history of English language teaching. Oxford: Oxford University Press.

Iwashita, N., May, L. & Moore, P. J. (2017) Features of discourse and lexical richness at different performance levels in the APTIS speaking test (Final report). The British Council/University of Cambridge Local Examinations Syndicate.

King, N. & Horrocks, C. (2010). Interviewing in qualitative research. London: Sage.

Levy, M. & Moore, P. J. (2018). Qualitative Research in CALL. Language Learning and Technology 22(2), 1-8.

Miles, M. B., & Huberman, A. M. (1994). Qualitative data analysis (2^nd ed.). Thousand Oaks, CA: Sage Publications.

Moore, P. J. (2013). An emergent perspective on the use of the first language in the EFL classroom. The Modern Language Journal 97(1): 239-253. DOI: 10.1111/j.1540-4781.2013.01429.x0026 7902/13/239–253

Moore, P. J. (2017). Unwritten rules: Code choice in task-based EFL learner discourse. In Fenton-Smith, B., Humphreys, P. & Walkinshaw, I. (eds.) English as a medium of instruction in higher education in the Asia-Pacific (pp. 299-320). Dordrecht: Springer. doi: 10.1007/978-3-319-51976-0_16

Moore, P. J. (2018). Task-based language teaching (TBLT). In Liontas, J. I. (ed.) TESOL encyclopedia of English language teaching. New Jersey: Wiley. doi: 10.1002/9781118784235.eelt0175

Moorkens, J. (2018). Machine translation: Evolution and quality expectations. Keynote presented at Google Translate & Modern Languages Education, University of Nottingham, UK, 29 June.

Murray, L., & Barnes, A. (1998). Beyond the “wow” factor—evaluating multimedia language learning software from a pedagogical viewpoint. System 26, 249-259.

Ortega, L. (2014). SLA for the 21st century: Disciplinary progress, transdisciplinary relevance, and the bi/multilingual turn. Language Learning, 63 (Supplement 1), 1–24.

Sauro, S., & Chapelle, C. (2017). Toward langua-technocultural competence. In C. Chapelle & S Sauro (Eds.), The Handbook of technology and second language teaching and learning (pp. 459-472). Hoboken, NJ: John Wiley & Sons, Inc.van Lier, L. (2004). The ecology and semiotics of language learning. A sociocultural perspective. Dordrecht: Klewer.

Way, A. (2018). Quality expectations in machine translation. In Moorkens, J., Castilho, S., Gaspari, F., and Doherty, S. (Eds.), Translation quality assessment: From principles to practice (pp. 159-178). Cham, Switzerland: Springer Nature.