Some remarks on stylistic shifts in interlanguage speech: The case of 'reading' and 'speaking' errors. (Linguistics).

1. Introduction

It is a widely accepted view in the area of second language acquisition that the L2 learner's interlanguage has the status of a linguistic system in its own right. Hence, it is characterised by features shared by all natural languages and, most importantly, by variability. Variability in interlanguage (IL) can be roughly defined as varying degrees of accuracy (relative to the relevant target language (TL) norm) in the learner's IL performance, or as diversity of non-TL forms (i.e. IL variants of TL targets) attested in IL production data. (1) The factors underlying IL variation are manifold in their nature and do not always lend themselves to straightforward classification and labelling. A specific issue that has frequently been addressed for the past three decades is if, and how, the learner's L2 performance varies systematically in stylistic shifts.

The notion of IL stylistic variation draws upon the foundations of the Labovian sociolinguistic paradigm, particularly his 'Observer's Paradox' (e.g. Labov 1970, 1972), the claim being that every speaker style-shifts in terms of linguistic (e.g. phonetic) parameters as the speech situation changes, and "styles can be ranged along a single dimension, measured by the amount of attention paid to speech" (Labov 1972: 208, original italics). The amount of attention paid to language form is assumed to be in direct proportion to the level of formality characterising the situation in which IL performance takes place, and, as stated by Major (1994: 190), "[a]ll other things being equal, the more formal the style, the more target-like the production". (2) The variable level of formality of style is concretised as a function of the variable experimental task in which IL production is elicited, with the tasks usually arranged on a scale ranging from free speech (least formal) through text reading, sentence reading, word list reading, to grammatical judgements (most formal) (e.g. Tarone 1983). The evidence to support the claim that free oral production indeed evidences less target-like variants and more native language (NL) interference than, e.g. text reading comes from a number of studies, such as Dickerson (1975), Tarone (1979), Major (1987) (see also Preston (1996) for a state-of-the-art statement on the variationist paradigm in Second Language Acquistion (SLA)).

The present paper, however, attempts to cast some doubt on the applicability of the stylistic variation paradigm, at least in the shape outlined above, to the context of second language production. It also reports on two studies examining the L2 pronunciation in Polish students of English. The first experiment aimed to assess the extent to which the subjects' IL production accuracy varied in two elicitation tasks, i.e. free speech and text reading. The second study involved impressionistic evaluation of the degree of foreign accent evident in those "reading" and "speaking" productions, with a view to determining if the IL speech generated in the former context did indeed give less of an impression of a foreign accent.

2. Study 1: the errors

2.1. Subjects

The subjects were 13 first-year students at the School of English (English philology), Adam Mickiewicz University in Poznan, Poland. They were aged between 18 and 21 and all had been learning English in the classroom setting, for about 6 years.

2.2. Methods

The data were collected in two elicitation tasks, i.e. text reading and free oral production. On the former occasion, the students read short passages from one of their English textbooks (250 words, on average). As for the spoken part, the subjects were asked to talk for a while about their overall impressions concerning their first year at university, which was slowly coming to an end at the time of the recording session. (3)

The recordings were transcribed phonetically by the present author, who relied primarily on auditory impressions, aided by some acoustic analyses conducted on digitised data, with the help of the Waves+ package.

2.3. Errors

The phonetic variables selected as indicators of the overall pronunciation accuracy were as follows:

(a) the vowels /e/, /**/, /[conjunction]/, /a:/, /D/, /**:/, /**/, /u:/ and /[contains as member]:/;

(b) the consonants /[theta]/, /**/, /[integral]/, /**/, /**/ and /**/;

The occurrence of each of the above tokenis in the students productions was labelled as either "target-like" or "non-target-like". The frequency of the "target-like" instances, relative to the number of all potential contexts, was then calculated for each of the two sound types, each of the two tasks, and for each of the 13 subjects.

2.4. Results

(a) vowels

The relative frequency of correct vowel tokens, in speaking as compared to reading, is presented in Figure 1.

Contrary to the pattern hypothesised within the stylistic variation paradigm, the raw numbers point to more target-like performance in speaking -- 43% of correct vocalic segments, compared with 40.1% in reading. However, since this difference is remote from significance (at p<.54), no true advantage of either elicitation task may be posited.

(b) consonants

The frequency of correct consonant realisations is presented in Figure 2. Here again, the numbers indicate an apparent advantage of speaking -- 59.4%, against 56.6% obtained for reading. This difference also falls much below the threshold of statistical significance, so we must limit ourselves to concluding that production accuracy did not systematically vary according to elicitation task.

2.5. Discussion

The corpus data did not provide support (although no significant invalidation, either) for the stylistic variation paradigm, wherein the reading task is predicted to evidence more target-like pronunciation than speaking. A few remarks about the plausibility of the variationist framework, outlined in Section I, are now in order.

Firstly, it is worth pointing out that there are a number of studies which do not corroborate the claim that there is a positive correlation between formality of the elicitation task and pronunciation accuracy in IL speakers. One such example is Sato (1985). Sato analysed the TL English speech of a Vietnamese boy, at 4 points during a 10-month period of observation. The data were collected in 4 different communication tasks: free conversation, text reading, and elicited imitation of words or phrases. The main focus of the study was the subjects' performance on word-final consonants and consonant clusters. For the former, significant task variation occurred in two (of the four) samples but not in the two others. Sato speculates that this result may be due to the fact that the boy's pronunciation was relatively target-like, in this respect, from the very beginning, and therefore might not have been as prone to stylistic variation. On the other hand, significant task variability for word-final clusters occurred in all samples but the first. Sato goes on to hypothesise that this lack of variation in the earliest sample may have been due to the overall low target-like performance on all tasks. It is difficult to disagree with Sato's reasoning, but if so, one is left to conclude that the claims about task variability pertain solely to intermediate learners, which substantially reduces the scope of applicability of the stylistic variation paradigm in the interlanguage context.

Major's (1992) study went a step further in challenging the stylistic continuum hypotheses, at least in the form outlined in Section 1. Major compared the performance of 4 Brazilian Portuguese speakers on TL English consonant clusters. The data were collected in 3 samples, in word-list reading and text-reading tasks, the former supposedly entailing a more formal style and, consequently, a more TL-like pronunciation than the latter. (4) However, contrary to this expected pattern, there were significantly more target-like occurrences of the variables in the text than in the word-list. Among other possible explanations of these somewhat puzzling results, Major proposes that the two tasks in question may not so much differ in formality as to expect large differences in production accuracy. This, again, would to some extent weaken the interlanguage stylistic continuum claims referred to in Section 1.

Major (1987) pointed to another problem inherent in viewing increasing TL production accuracy as a function of increasing formality of style, since "extra-linguistic factors may come into play, such as stress, which may produce poor performance even in formal styles" (Major 1987: 121, note 8). In this case pronunciation may actually be more error-prone in formal than in casual situations.

It also has to be borne in mind that a reading task may be more conducive to certain types of pronunciation errors than free speech, due to the enhanced effects of orthography. As noted by Markham (1997: 102), "orthographic information ... may cause increased monitoring of pronunciation -- for better or worse".

3. Study 2 -- evaluations

3.1. Methods

Excerpts of 15-20 seconds in length were extracted from the recordings used in Experiment 1. Care was taken to choose strings of speech which were relatively free of syntactic deviations. There were 26 excerpts (13 subjects, two elicitation tasks) derived from the students' productions, plus 6 additional samples, obtained from 3 control subjects -- native speakers of British English. The samples were randomised (although the two tasks were always kept apart) and then presented to the judges for assessment of the degree of foreign accent. The scale ranged from 1 to 5, with 1 standing for "very strong foreign accent" and 5 signifying "no accent at all; definitely native". (5)

3.2. Listeners

The following three groups of judges were asked to evaluate the accentedness of IL speech samples:

(1) Polish teachers from the School of English (henceforth POL);

(2) British English native speakers, teachers at the School of English, resident in Poland for at least 5 years (PNS);

(3) 'pristine' native speakers of English, permanently living in Buckinghamshire, UK (ENG).

3.3. Results

As can be deduced from Figure 3, no significant difference between the evaluations of reading versus speaking performance was observed. This is true both for the joint results (mean scores of 2.39 and 2.50, respectively), and for two out of the three judge groups considered individually. This single exception were the PNS listeners, who turned out to be significantly more generous with points in the case of speaking than reading (2.51 and 2.17, respectively; p<.02), which is illustrated in Figure 4.

3.4. Discussion

Any investigation of IL performance, attempting to quantify the degree of phonetic deviation from TL norms, may rely on statistical analyses of pronunciation errors, or on impressionistic judgements of foreign accent passed by native speakers. However, even the most meticulous instrumental analysis of the speech signal will invariably miss some crucial, albeit elusive, constituents of "sounding foreign", and will fail to provide a reliable indicator of the "nativeness" of one's speech. Therefore, the ultimate measure of TL accuracy must be the native speaker and his impressions (however subjective these may be), rather than pure figures denoting the frequency of various errors, although it stands to reason that the former must in some way be affected by the latter.

As for the relative TL accuracy of reading or speaking, measured in terms of accent evaluations, the present study did not attest more target-like performance on the task that allowed higher degrees of monitoring, i.e. text reading. In fact, the reverse pattern could be observed, at least in the case of PNS judges. One could explain this apparent superiority of speaking by means of a possible relaxation of the judgement criteria adopted by the listeners in the case of a more demanding elicitation task. That is to say, the judges might lower their expectations (and, consequently, the yardstick they used in foreign accent evaluation) when exposed to IL performance in a task posing a substantial processing load on the learner, i.e. free speech, as compared to "simple" text reading. However, a more plausible explanation, at least in the case of studies investigating the speech of (not necessarily thoroughly schooled) immigrants, is that the subjects may not have had much opportunity to practice and develop brilli ant reading skills in the second language. Therefore their performance on any task that involves reading aloud a text in front of an audience, might be actually poorer than that evidenced in the good, old, familiar "free speech" exercises. Besides, when asked to evaluate the nativeness, i.e. naturalness of a learner's pronunciation, native speakers may also take into account, subconsciously rather than consciously, the naturalness of the speech context itself. Needless to say, reading aloud can hardly count as a neutral, unmarked context, which one would typically associate with native speaker productions. Thus, reading a text (let alone a list of words or sentences) may inherently be stigmatised as foreign-sounding, and therefore produce the impression of less target-like performance, than a speaking exercise. Why all the above factors, speculated to disqualify reading as a task more conducive to target-like performance than speaking, should be especially operative in the case of the PNS judges, remains yet to be established.

Evidence suggesting an advantage of reading over speaking is rather scarce in the SLA literature pertaining to assessments of foreign accent by native speakers. In Thompson's (1991) experiment, for example, sentence reading was rated as more accented than passage reading, which, in turn, ranked higher on the accentedness scale than spontaneous speech. Needless to say, this ordering of tasks is a mirror image of the one predicted within the Labovian framework. Some limited support for the reading-outperforms-speaking hypothesis comes from the study by Bongaerts, Planken and Schills (1995). Their subjects, both experienced and inexperienced Dutch learners of English, were tested in 4 elicitation tasks: (1) free speech; (2) text reading; (3) sentence reading, and (4) word list reading. The evaluation of the degree of accentedness revealed that, indeed, the scores on Task (4) were much higher than those on the remaining three tasks. However, in the case of the inexperienced group, it was sentence reading (task 3) , and not free speech (1), that generated the least target-like performance. Thus, the ordering of tasks according to phonetic accuracy they generate, differs again from the one anticipated by the variationist paradigm.

4. Conclusion

The results of the empirical studies which examine the variable phonetic accuracy as a function of stylistic shifts, are at best inconclusive. The hypothesis that more formal tasks trigger more attention to form, and thus more target-like performance, rests on rather shaky assumptions.

One of the most disputable matters that present themselves in the context of applying the Labovian framework to SLA, regards the very notion of relating different elicitation tasks to varying levels of formality. What is it, for instance, in the nature of text-reading that produces a more formal style than a free-speech task, if on both occasions the subjects are talking to the same tape recorder, in the presence of the same researcher, so the speech situation variables are held constant? The attempts to define the relative formality of style as a function of the relative formality of the speech situation have also been accused of circularity (e.g. Dewaele 1995). This is evident in the following line of reasoning: formal style is the style used in a formal situation, and we know a given situation is formal because the style employed by the speaker is formal, as well. (6)

There are also serious difficulties inherent in linking task-related style shifting with shifts in the degree of attention the speaker pays to language form. As Tarone (1988: 41ff), herself an advocate of the Labovian paradigm, remarked, a psychological phenomenon like "attention to speech" is impossible to verify empirically. It is not feasible to prove that, for instance, on a word-list reading task the speaker is more attentive to form than on telling a story, as "the workings of the mind do not lend themselves to empirical observation" and "individuals' attention may waver from one minute to the next during a task" (Tarone 1988: 42). Besides, it is highly controversial, or plainly oversimplified, to claim that, e.g. in free speech tasks the IL learner pays the minimum amount of attention to form, in favour of the context. This may indeed be true for native speakers of a language (for whom the Labovian framework was originally designed), whose speech production processes are fully automatised and do not re quire conscious control. In the case of L2 learners, a substantial degree of attention is always necessary, even, or above all, on speaking tasks. As observed by Sato (1985: 195), "this attention must be paid, not simply to language form but also to other demands on real-time discourse production: ... encoding of rhetorical structure, lexical items, clause sequencing, etc. Phonological phenomena ... such as consonant clusters appear to rank low on this list of demands on the learner". In other words, in elicitation tasks like 'free speech' a great deal of attention is required, but this attention has to be divided among various subcomponents of language form: phonetic accuracy, syntactic correctness, vocabulary selection, as well as the content, which may lead to a processing overload. Hence, one should perhaps talk about diffusion of attention in speaking exercises, since the claim positing virtual lack of attention paid to language form on such tasks seem to be an ungrounded oversimplification.

[FIGURE 1 OMITTED]

[FIGURE 2 OMITTED]

[FIGURE 3 OMITTED]

[FIGURE 4 OMITTED]

(1.) In fact, James (1990) insists on a distinction between the two definitions of the term presented above, calling the former variation and the latter variability (proper), which is understood as contextual variability. In the case of variation, "the contextual influence proceeds from context-free to context-sensitive", whereas with variability "it proceeds from 'non-systematic' to 'systematic', evidencing the establishment of stable 'form-function' relationships of units." (James 1990: 239). This distinction, however, is generally ignored in the literature on the subject, with the two terms being used interchangeably, as in e.g. Sato 1985, Tarone 1988.

(2.) However, as reported by e.g. Beebe 1987, the reverse may occur if a given phonetic variable carries some social value in the learner's native language. Then, this prestige NL variant may be rampantly used precisely in formal situations (i.e. in the sociolinguistically appropriate setting), resulting in less target-like performance than in informal style.

(3.) The students were actually recorded both at the beginning of the academic year (October) and at the end (May); for the sake of brevity, the paper considers the May results only.

(4.) The data also included free speech samples, but these were disregarded in the statistical analysis, as too few tokens of the investigated items appeared in those sections.

(5.) The evaluation procedure was modelled on that adopted by Bongaerts, Planken and Schills (1995).

(6.) Dewaele (1995) proposes an independent measure of the formality of style, involving the relative implicitness or explicitness of speech. This criterion, however, is rather inapplicable to the 'reading' versus 'speaking' context.

REFERENCES

Bayley, Robert -- Dennis, R. Preston (eds.)

1996 Second language acquisition and linguistic perspective. Amsterdam: John Benjamins.

Beebe, Leslie M.

1987 "Myths about Interlanguage Phonology", in: Georgette Ioup -- Steven H. Weinberger (eds.), 165-175.

Bongaerts, Theo -- Brigitte Planken -- Erik Schils

1995 "Can late starters attain a native accent in a foreign language? A test of the Critical Period Hypothesis", in: David Singleton -- Zsolt Lengyel (eds.), 30-50.

Dewaele, Jean-Marc

1995 "Style-shifting in oral interlanguage. Quantification and definition", in: Lynn Eubank -- Lary Selinker -- Michael Sharwood Smith (eds.), 233-240.

Dickerson, Lonna

1975 "The learner's interlanguage as a system of variable rules", TESOL Quarterly 9, 401-407.

Eubank, Lynn -- Lary Selinker -- Michael Sharwood Smith (eds.)

1995 The current state of interlanguage. Amsterdam: John Benjamins.

Fisiak, Jacek (ed.)

1990 Further insights into Contrastive Analysis. Amsterdam: John Benjamins Publishing Company.

Gass, Susan M. -- Carolyn, G. Madden (eds.)

1985 Input in Second Language Acquisition. Rowley, Ma: Newbury House Publishers.

Ioup, Georgette -- Steven H. Weinberger (eds.)

1987 Interlanguage phonology: The acquisition of a second language sound system. Cambridge, Ma: Newbury House Publishers.

James, Allan R.

1990 "Phonological structure and the contextual variability of interlanguage segments", in: Jacek Fisiak (ed.), 231-240.

Labov, William 

1971 "The study of language in its social context", Studium Generate 23: 30-87.

1972 Sociolinguistic patterns. Philadelphia: University of Pennsylvania Press.

Leather, Jonathan -- Allan, R. James

1992 New Sounds 1992. Amsterdam: University of Amsterdam.

Major, Roy C.

1987 "A model for interlanguage phonology", in: Georgette Ioup -- Steven H. Weinberger (eds.), 101-124.

1992 "Transfer and developmental factors in second language acquisition of consonant clusters", in: Jonathan Leather -- Allan R. James (eds.), 128-135.

1994 "Current trends in interlanguage phonology", in: Mehmed Yavas (ed.), 181-203.

Markham, Duncan

1997 Phonetic imitation, accent, and the learner. Lund: Lund University Press.

Preston, Dennis R.

1996 "Variationist perspectives on second language acquisition", in: Robert Bayley -- Dennis R. Preston (eds.), 1-45.

Sato, Charlene J.

1985 "Task variation in interlanguage phonology", in: Susan Gass -- Carolyn G. Madden (eds.), 181-196.

Singleton, David -- Zsolt Lengyel (eds.)

1995 The age factor in second language acquisition. Clevedon: Multilingual Matters Ltd.

Tarone, Elaine

1979 "Interlanguage as chameleon", Language Learning 29, 181-191.

1983 "On the variability of interlanguage systems", Applied Linguistics 4, 142-163.

1988 Variation in Interlanguage. London: Edward Arnold.

Thompson, Irene

1991 "Foreign accents revisited: The English pronunciation of Russian immigrants", Language Learning 41, 177-204.

Yavas, Mehmed (ed.)

1994 First and second language phonology. San Diego: Singular Publishing Group, Inc.

*[Text unreadable in original source]