Three-word sequences

This paper focuses particularly on three-word sequences (e.g. I don't know, I would like). The unit of three words per sequence was chosen because three-word units are embedded with sufficient contextual information regarding functional language use, as has been shown in previous studies (e.g. Adolphs & Durow, Reference Adolphs, Durow and Schmitt2004; Lin, Reference Lin and Romero-Trillo2013). Empirical research also shows that the majority of high-frequency core phrases in written discourse are three-word phrases (Simpson-Vlach & Ellis, Reference Simpson-Vlach and Ellis2010). These three-word phases often form the basis of the larger sequences of four or more words and may be acquired or processed as parts as well as wholes (Conklin & Schmitt, Reference Conklin and Schmitt2008). In addition, three-word sequences are analytically more manageable: analysing two-word sequences (e.g. of the, to be) would include too many phrasal verbs and grammatical colligations, while considering a larger unit in the sequence (e.g. at the end of the) would reveal too few examples.

The keyness approach: Analysing key sequences

With a view to examining the development of three-word sequences during the online interaction, the items frequently used by Taiwanese participants were analysed to examine their approximation to those sequences used by native English-speaking participants. In doing so, this study employed a corpus linguistic approach using Wordsmith Tools 5.0 (Scott, Reference Scott2008) to identify the key sequences, which refer to those sequences that occur unusually frequently or unusually infrequently in a text as compared with some kind of reference corpus (O'Keeffe et al., Reference O'Keeffe, McCarthy and Carter2007; Scott, Reference Scott, McCarthy and O'Keeffe2010). The procedure works by comparing the observed frequency of each three-word sequence in the target corpus with its equivalent in the reference corpus. For this study, the Taiwanese and British participant datasets were set as the target and reference corpus respectively, working on the basis of the relative frequencies of each three-word sequence in the two datasets and further identifying the sequences with significant unusual frequencies in the Taiwanese participants’ discourse; accordingly, the key sequences are generated and ranked by keyness, which the software calculates by applying the log likelihood (LL) test of statistical significance to the difference between the target and reference corpus frequencies for each item.

Although previous studies have extensively employed keyword analysis in the investigation of different contexts of language use (e.g. Baker, Reference Baker2006; Bondi & Scott, Reference Bondi and Scott2010; Harvey, Reference Harvey2012), few, if any, have applied this method for understanding language development online. This present study thus intends to demonstrate the merit of keyness analysis as a tool for examining the level of approximation with regard to the use of three-word sequences by Taiwanese and British participants.

Key three-word sequences

Using the Keyword function of Wordsmith Tools 5.0, for p < .01, the cut-off LL value of 6.64 reveals 73 unusually infrequent three-word sequences used by Taiwanese learners, as compared with British participants’ discourse. Table 1 illustrates the items that were commonly used by the British participants. As can be seen, most of these were not used by the Taiwanese learners. In this list, it can be noted that the three-word sequences based on the coordinating conjunction and, such as and go to, and I love and me and my, occurred in the native speakers’ discourse frequently, while no instances of these three sequences can be found in the Taiwanese pupils’ discourse. Also, the sequences with the modal would, such as I would love, would love to and it would be commonly used by natives are not found in the Taiwanese learners’ texts either. A further investigation of the context indicates the high frequency with which modal would was used by British students to respond to requests in online communication, as these are generally perceived as more polite and less forceful.

Table 1:

List of underused three-word sequences in the Taiwanese datasetc

*LL > 10.83 indicates p < .001.

Decline of key sequences over time

Table 2 presents the number and percentage of three-word key sequences that were identified based on each phase of Taiwanese discourse in comparison with the sequences used by British participants. For p < .01 with 1 d.f., 6.63 as the cut-off of LL value, 76 (15.51%) and 65 (13.10%) items were identified as key out of the total 490 and 496 sequences for Phase 1 and 2 respectively. This figure then dramatically decreases to 32 (4.87%) in the last phase of the programme. This continuous decline in the number and percentage of key sequences indicates that the Taiwanese participants gradually used fewer three-word sequences that are identified as statistically significantly different relative to those used by the British participants.

Table 2:

Number and percentage of key sequences in Taiwanese dataset

The decline of key sequences shows an increasing convergence between the use of three-word units by Taiwanese and British native speakers of English. This confirms previous findings (Adolphs & Durow, Reference Adolphs, Durow and Schmitt2004; Crossley & Salsbury, Reference Crossley and Salsbury2011; Qi & Ding, Reference Qi and Ding2011) that intercultural contact with native speakers of English fosters the longitudinal development of the use of multi-word sequences, and this also contributes additional evidence that online interaction can achieve such positive results even with adolescent EFL learners at a beginner level. The method here, which focuses on naturally occurring language output online, diminishes the effects of the artificial contexts often created in language testing settings.

While the findings indicate that online exchanges with native speakers show an increase of approximation with the British participants in terms of the use of multi-word sequences, other factors influencing such approximation also need to be considered. First, the decrease of key sequences over time may be associated with the topic shifts; that is, the participants initially talked more about their own cultures and the differences between them (e.g. the different food cultures), but as time went on they discussed shared topics in order to build friendships and maintain relationships. Another possible reason for the approximation between the two groups can be attributed to the fact that some correspondence extracts show Taiwanese learners occasionally imitated British native speakers, in which case the learner may have control of part of the holistic or the componential meaning (Wray, Reference Wray2002). Crystal (Reference Crystal2006) also observes that in online communities, although members come from different backgrounds and write in different styles, they tend to accommodate each other, and as such their contributions progressively develop a shared linguistic style.

Another issue that needs to be stressed in this discussion is that there is debate as to whether native-speaker models are the most appropriate basis for language learners. From the perspective of World Englishes or English as a Lingua Franca (ELF) (e.g. Jenkins, Reference Jenkins2009; Rajagopalan, Reference Rajagopalan2004), learners who use specifically native-speaker norms may well risk adopting a false identity. Although there has been considerable debate on this issue in EFL contexts, learners still need models of some kind as a point of reference, particularly novice learners (Cullen & Kuo, Reference Cullen and Kuo2007; Lin, Reference Lin2015). Research on students’ perspectives shows that learners across a diverse range of countries and contexts of language use regard the approximation to native-speaker English as a main goal in the language learning process (e.g. Timmis, Reference Timmis2012). This notwithstanding, I would argue that raising EFL learners’ awareness of the multi-word units commonly used by native speakers of English does not necessarily mean training our learners to be like them. The aim is to make learners aware of how human beings interact with each other and how they create and sustain successful interactions. To that end, those sequences that commonly occurred in authentic communication are crucial for language learning, and it is upon those language patterns that we should focus in EFL pedagogy.