Extraction of Predetermined Pragmatic Features Introduced in the Corpus- Corpus-Based ELT Textbooks from the NICT JLE Corpus (Miura, 2009)

Miura (2009) conducted her initial study investigating the pragmatic features in learner corpora. The purpose of the study was to show pedagogically a gap between the expectations of textbook writers and editors and the actual language use of Japanese EFL learners in the NICT JLE Corpus. The author focused on the pragmatic or

105

conversational features retrieved from the frequency lists of a large corpus in corpus-based ELT textbooks called the Touchstone (McCarthy & Carter, 2005a; 2005b; 2005c;

2006a; 2006b). This series was edited based on research into the Cambridge English Corpus in order to present natural language in authentic texts (McCarthy, 2004; O’Keeffe, McCarthy, & Carter, 2007). Each book of the series features so-called conversation strategies, composed of the frequent chunks or multi-word strings retrieved from the corpus, such as hedging, vagueness, discourse marking, the preservation of face, and the expression of politeness, which have pragmatic functions (McCarthy & Carter, 2006;

O’Keeffe et al., 2007). A total of 81 different items of conversation strategies with the description of each function are introduced in books for beginners or CEFR A1 learners (i.e., Book 1), high beginners at CEFR A1 and the entry level of A2 (i.e., Book 2), low-intermediate learners at CEFR A2 and the entry level of B1 (i.e., Book 3), and intermediate learners or B1 learners (i.e., Book 4). In this study, Miura (2009) predetermined the following pragmatic features to be investigated in the whole data of the NICT JLE Corpus across nine different proficiency levels, containing 1,281 learners with 47,900 types and 1,762,919 tokens (p. 142): I mean, I guess, really, just, maybe, actually, kind of, like, and so. The features were all automatically extracted from the corpus, without any contextual references and task-effect considerations. The overall results indicated that these pragmatic features were rarely produced by basic learners, but tended to be produced more frequently by intermediate learners than upper intermediate and advanced learners. The rough observation of these features in concordance lines suggested that most of them functioned as fillers when “searching for the appropriate expression” (see Fung & Carter, 2007) and “denoting thinking process” (see Müller, 2005). The learners’ use of these features were rather different from what was explained in the textbooks, probably because the learners manipulated their communication

106

strategies for the negotiation of meaning in order to compensate for their lack of command of the language, or “their lack of vocabulary and productive skills” (Miura, 2009, p. 152).

Therefore, it was indicated that intermediate learners had more frequent productions possibly due to their “verbosity” (Faerch & Kasper, 1989), or “the waffle phenomenon,”

which Edmondston and House (1990) defined as the “excessive use of linguistic forms to fill a specific discourse ‘slot’ or ‘move,’ i.e. [to] achieve a specific pragmatic goal” (pp.

273-274).

3.3.2 A “Form-to-Function” Analysis of Discourse Markers with Multi-Functionality in Different Interactional Situations (Miura, 2011; 2014)

Following the initial study, in order to observe learner-proficiency differences in the use of pragmatic features, the author examined the use of discourse markers such as actually and I guess (Miura, 2011), and well, I mean, kind of, and like (Miura, 2011;

2014) in the NICT JLE Corpus with a comparison of native speakers’ data, following the classificatory descriptions of discourse markers made by Fung and Carter (2007) and Müller (2004; 2005). As previously mentioned, the NICT JLE Corpus is composed of written transcripts of the SST, which is divided into five different stages. In order to overcome the weakness of the research methodology in the initial study (Miura, 2009), which did not take into account the interactional differences of speech production, the author, in the next studies (2011; 2014), divided the learner data into three different sub-corpora: (i) monologues, where learners were asked to describe the pictures given (including Stages 2 and 4), (ii) casual dialogues, where learners had casual conversations with the interlocutors (including Stages 1 and 5, and follow-up sessions of Stages 2, 3, and 4), and (iii) role-play dialogues, where learners were instructed to conduct role plays simulating particular social situations such as shopping (including Stage 3). Although the

107

initial study (Miura, 2009) did not investigate the functions of pragmatic features, the subsequent studies (Miura, 2011; 2014) aimed to conduct a “form-to-function” analysis, concerning not only the interactional effects on learner production, but also the classification of the multi-functionality of discourse markers. First, Miura (2011; 2014) resorted to Fung and Carter (2007)’s classification of discourse markers into four functions: interpersonal (e.g., kind of, like, well, you know, I see, etc.), referential (e.g., and, because, but, however, so, etc.), structural (e.g., and, finally, first, now, well, etc.), and cognitive (e.g., and, like, I mean, well, you know, etc.). For example, the procedure of the analysis of well was as follows: all the occurrences of well were retrieved on the basis of the filler tags with which the corpus was already annotated, which allowed the author to exclude the adverbs, adjectives, and nouns of well. It was observed that well was the most frequently produced by learners at SST Level 8 in role-play dialogues, followed by Level 9 in monologues, but infreqeuntly produced by learners from Levels 3 to 6, as well as by 40 native speakers, in all three situations. The frequent production by intermediate and upper intermediate learners was again assumed to be attributed to their tendency of verbosity. Thus, the role-play dialogues produced by Level 8 learners contained approximately 1,800 occurrences of well per 100,000 tokens (i.e., approximately 1,300 raw frequencies among 70,404 tokens). According to Fung and Carter (2007), there are mainly three different functions of well: interpersonal (i.e.,

“indicating attitudes” when the speaker cannot answer either yes or no), structural (i.e.,

“opening and closing of topics” when the speaker wants to change the topic), and cognitive (i.e., “denoting thinking process” when the speaker needs some time before producing the following utterance) (p. 418). However, due to the vast numbers of extracted target features, detailed functional analyses of the aforementioned features were not practically possible.

108

Further, Miura (2011; 2014) preliminarily attempted to conduct a “form-to-function” analysis, by observing like, only focusing on the data of learners at Levels 3, 6, 9 and native speakers. Based on the notion of “optionality,” which means that “[discourse markers] are semantically and grammatically optional” (Fung & Carter, 2007, p. 414) and that “they are syntactically optional and contribute little or no propositional meaning to the utterance that contains them” (Müller, 2004, p. 1158), Miura (2011; 2014) identified the discourse marker, like, in the following procedures: (i) retrieved all occurrences of the lexical item, like; (ii) manually eliminated the forms of the lexical verb and preposition, like, from the target data; (iii) identified the discourse markers, like, when they were

“optional” and had functions such as “searching for the appropriate expression” (Müller, 2005, p. 208), “making an approximate number or quantity” (p. 210), “introducing an example” (p. 212), “introducing an explanation” (p. 215), and “marking lexical focus” (p.

219).

In contrast with well, like was the most frequently produced by native speakers in casual dialogues (i.e., 1,600 occurrences per 100,000 tokens), about 1.8 times more than that produced by Level 9 learners, 5 times more than that produced by Level 6 learners, and 17 times more than that produced by Level 3 learners. In contrast, in role-play dialogues, approximately 400 normalized frequencies per 100,000 tokens were produced by native speakers and Level 9 learners, while in monologues, 200 frequencies were produced by these groups. The results indicated that like tended to be used in more casual conversations especially by advanced learners and native speakers. A detailed functional analysis was again not possible because it was difficult to match the forms and functions of the discourse markers due to their multi-functionality. For example, in “Like I can get there in like ten minutes or so” (Miura, 2014), each like can have more than one function according to the aforementioned definitions given by Müller (2005), which made

109

it difficult for the author to determine the functions, since the author had no access to the speakers to confirm the intentions of their utterances.

To summarize, Miura (2009; 2011; 2014) conducted the aforementioned corpus pragmatic studies in order to examine the pragmatic competences of Japanese EFL learners at different proficiencies, focusing on the pragmatic features or discourse markers at the surface level. However, the series of studies indicated the difficulties of conducting “form-to-function” analyses, or identifying and analyzing the functions based on only the extracted surface forms of predetermined linguistic features retrieved from the concordance lines.

3.4 Analyses of Pragmatic Functions in Longer Stretches of Discourse: Extraction of Manually Annotated Requestive Speech Acts

In the next stage, the author attempted to explore the possibilities of expanding the scope of spoken learner corpora from investigations of surface forms (e.g., the lexico-grammatical features of discourse markers) to those of pragmatic functions (e.g., speech act expressions) (Miura, 2015a; 2015b; 2015c; 2016a; 2016b; 2017; Miura

& Sano, 2014). Rather than automatically extract the linguistic items manifesting pragmatic functions from the corpus and then attempt to match the forms and functions (Miura, 2009; 2011; 2014), the author instead attempted to conduct a “function-to-form”

analysis, in order to examine the development of pragmatic competences and to explore the pragmatic criterial features distinguishing the different levels of proficiency. The author manually annotated requestive speech acts, drawing on and revising the CCSARP coding scheme devised by Blum-Kulka et al. (1989). Rather than investigate the whole data, the author extracted only the learner utterances in role-play sessions, where the interlocutor and learner interacted dialogically.

110

The current section describes how the author attempted to identify and annotate speech act realizations as part of the founding studies to the current doctoral study (see Table 3.3). First, the preliminary studies are reviewed as the pilot annotation of small-scale data using XML tags (Miura & Sano 2014; Miura, 2015a). Second, the development of multi-layered annotation schemes for larger data with a tool called UAM CorpusTool (UAMCT) (O’Donnell, 2012) is reviewed (Miura, 2015b; 2015c; 2016a;

2016b). Thus, Miura (2015b; 2017) attempted to explore the possibility of assessing learners’ sociopragmatic competences and annotating the degree of politeness in the identified pragmalinguistic features of requests.

Table 3.3

A list of the author’s previous studies regarding requestive speech acts in the NICT JLE Corpus

Publication Date

Annotation

Schemes Task Task Types

The No.

of Subjects

Proficiency Levels

Miura (2015a)

Requestive speech acts

Shopping

Basic 23 SST Levels

1, 2, & 3 Intermediate 20 SST Levels

4 & 5

Advanced 40 SST Levels

6, 7, 8, & 9

Train

Basic 10 SST Levels

1 & 2 Intermediate 30 SST Levels

3, 4, & 5

Advanced 40 SST Levels

6, 7, 8, & 9 Miura

(2015b)

1. Requestive

Speech acts Shopping Beginner &

Intermediate 67 CEFR A1 (i.e., SST Level 3)

111 2. Politeness

Intermediate 67

CEFR A2 (i.e., SST Levels 4

& 5)

Advanced 66

CEFR B1 (i.e., SST Levels 6 to 8)

Beginner, Intermediate,

& Advanced

14 Native speakers

Miura (2015c;

2016a)

1. Requestive speech acts 2. Requestive

functions 3. Requestive

naturalnessⁱⁱⁱ

Shopping

Beginner &

Intermediate 68 CEFR A1 (i.e., SST Level 3)

Intermediate 114

CEFR A2 (i.e., SST Levels 4

& 5)

Miura (2017)

1. Requestive speech acts 2. Requestive

functions 3. Requestive

politeness

Shopping

Beginner &

Intermediate 68 CEFR A1 (i.e., SST Level 3)

Intermediate 114

CEFR A2 (i.e., SST Levels 4

& 5)

Advanced 66

CEFR B1 (i.e., SST Levels 6 to 8)

3.4.1 XML Annotations of Requestive Speech Acts in the NICT JLE Corpus (Miura,