Flipping the script: A Corpus of American Television Series (CATS) for corpus-based language learning and teaching

Stefanie Dose
Department of English, Justus Liebig University Giessen

Abstract

1. Introduction: Spoken language in the EFL classroom

Corpus-based research has brought to light fairly extensive and empirically sound descriptions of ‘spoken grammar’ or a ‘grammar of conversation’ (e.g. Biber et al. 1999; Carter and McCarthy 2006) in the past couple of decades. So far, however, such insights on the lexico-grammatical differences between speech and writing are rarely considered in actual teaching and learning contexts, and mainstream EFL teaching still tends to focus on the written language (cf. Thornbury 2005: 34; Timmis 2005: 117). This is rather unfortunate, since oral language proficiency has officially become a major goal of modern communicative language teaching. Regional and supraregional curricula in Germany, for instance, now give equal importance to each of the four skills, i.e. listening, speaking, reading and writing (cf. Taubenböck 2007). Such developments are by no means restricted to Germany’s educational policy, which is, in fact, to a large extent based on the Common European Framework of Reference for Languages (CEFR; Council of Europe 2001). The CEFR stresses the importance of communicative skills and specifies target competencies in areas which directly concern features of spoken grammar, such as in the description of the competence levels in the categories ‘Functional competence’ and ‘Interaction strategies’. To master the skill of speaking and interacting in conversation, a speaker needs to take account of the differences between spoken and written language and be able to handle the communicative circumstances of conversation in a natural and idiomatic way.

In practice, the specified target competencies concerning spoken interaction seem to be very difficult to achieve even for advanced learners. The current status in Germany’s EFL classroom is rather disappointing: Even advanced students display major shortcomings in idiomatic, spontaneous spoken language use (cf. Kieweg 2000: 8; Kurtz 2001: 14; Mukherjee 2004: 247; Mukherjee 2009: 205). This most likely results from a) a lack of exposure to spontaneous spoken language in natural settings, b) a lack of instruction about the differences between spoken and written language, and c) a lack of speaking practice, which hinders the students from turning their knowledge into skills. It is especially at point b) where the applied corpus-linguistic community offers its assistance.

In order to help learners gain an awareness of the various qualitative and quantitative differences between spoken and written language and improve their oral competence, applied corpus linguists have advocated the use of spoken corpora in the classroom, e.g. with corpus-based teaching and learning materials but also with other data-driven approaches such as classroom concordancing (e.g. Braun and Chambers 2006; Mauranen 2004; Zorzi 2001). Attempts at bringing naturally occurring spoken language into the classroom remain rather scarce, however. One major problem is that linguistic corpora, especially spoken corpora, can be very difficult to handle for both teachers and learners. Large masses of anonymous data have much more appeal to the researcher than to the participants in a language class, who struggle with the seemingly unstructured, messy language bits they encounter in a spoken corpus. Acknowledging the difficulties that real language corpora pose in a pedagogical setting, researchers such as Braun (2005, 2006, 2007b) have recommended the design of so-called “pedagogically relevant corpora”, i.e. corpora which are specifically tailored to the practical needs and interests of teachers and students in terms of contents, size, format of the data and annotation. One prime example of a pedagogically relevant spoken corpus is the ELISA corpus, which consists of the transcripts of 28 videotaped interviews with native speakers of different varieties of English, who talk about their professional careers (see Braun 2004 and Braun 2006 for a detailed description of the project). To date, ELISA is still quite exceptional in this area. However, Römer (2008: 124) envisages a promising future for pedagogically motivated corpora, confirming a prediction made by Aston (2000: 16), who claimed that “language pedagogy is increasingly designing its own corpora according to its own criteria”.

2. CATS – A Corpus of American Television Series

2.1 Corpus design

CATS (Corpus of American Television Series) was compiled as an attempt to expand the range of pedagogical corpus resources. [2] The corpus consists of dialogues taken from four contemporary American television series which were selected according to a list of predefined criteria, having to do with the language used, the topics of the shows, the format/genre, and their popularity on U.S. American and German television. The following shows fulfilled the criteria to the greatest extent possible: Gilmore Girls (2000–2007; Season 4), Monk (2002–2009; Season 1), Six Feet Under (2001–2005; Season 1), and Veronica Mars (2004–2007; Season 1). [3] All of these are very successful productions in the tradition of ‘drama series’ or ‘dramedies’, targeted at adolescent and adult audiences. They have a limited cast and a relatively self-contained plotline in each episode of 40–50 minutes. Seven consecutive episodes were used from each of the four shows, totaling 28 episodes.

CATS consists of 160,592 words of transcribed language spoken by American English native speakers (with only a few exceptions). The four shows contribute in unequal parts to the corpus (see Table 1) because the word count per episode varied greatly from series to series. For example, Gilmore Girls has a much higher word count than the other series owing to the extremely quick dialogues for which the show is so well-known. The slightly varying length of the episodes also played a role. This ‘imbalance’ was not seen as problematic for the potential uses of the corpus, however, and it was considered more important to include the same number of complete episodes (with completed plotlines).

CATS contains mainly face-to-face dialogue between family members, friends, classmates, and colleagues, but there is also a small portion of monologic spoken language such as speeches, teacher talk, voice-overs, etc. Although the main focus of this study is on conversation as the most basic and typical kind of speech (cf. Hughes 2002: 13), the other types were not excluded because they were deemed useful models for EFL learners, too.

2.2 Transcription and annotation

I took advantage of two different websites offering transcripts – not scripts – of the individual episodes in html-format (www.twiztv.com and http://episodeguides.blogspot.com/, last accessed 30 Aug. 2010). [4] The transcripts are freely available for download and can easily be converted into text-files for further editing. They may be used for entertainment, teaching and other educational uses. However, it is crucial to note that the transcriptions on these websites are written by fans of the shows, who are usually non-linguists. They transcribe their favorite shows only for their personal fun and entertainment and then share their work with other fans of the shows on Internet platforms. As it turned out, they often transcribed inconsistently, missed details and added information which would be superfluous in a linguistic corpus. Furthermore, the transcriptions – despite the guidelines offered by the websites – were quite idiosyncratic. Since the inconsistencies, omissions, and inaccuracies by the lay transcribers greatly affected precisely the phenomena that researchers of spoken grammar might be interested in, namely items such as backchannels, fillers, repeats, discourse markers, etc., the transcripts had to be proofread and corrected very thoroughly. It was also necessary to unify the spelling options (e.g. OK vs. okay) to produce one coherent standard before the corpus was annotated and prepared for analysis. In sum, the transcripts from the websites were taken as useful ‘first drafts’ and subsequently edited extensively to meet the requirements of a corpus for educational purposes.

The spoken texts in CATS are represented by a very close orthographic transcription. [5] The transcription does not conform to any established transcription standard, but is customized to the needs and purposes of this project. Performance phenomena, backchannels, and other verbal expressions typical of conversational language have consistent orthographic representations, which are all recorded in a ‘transcription codebook’ to guarantee uniformity throughout the corpus. The transcription of performance phenomena was considered necessary because they are a characteristic part of spoken grammar and thus one focus of future analyses with this corpus. After all, their presence/absence can indicate how similar the language in television series is to naturally occurring speech. [6] The transcription also includes conventional punctuation symbols such as hyphens, periods, commas, question marks, and exclamation marks in order to reflect the typical intonation contours and syntactic boundaries associated with these symbols. While it is clear that this practice indirectly suggests written norms and is not the most common way of transcribing for a spoken corpus, punctuation marks make the interpretation of the transcripts much easier, and so it seemed to be the best solution for the anticipated purposes, i.e. lexico-grammatical analyses as well as classroom use by non-linguists.

A complex annotation scheme was considered unnecessary since it would not be required for future predominantly lexico-grammatical analyses. For instance, I refrained from annotating overlaps and length of pauses or any phonetic information (e.g. stress, intonation). Some basic annotation relating to speaker names, scene information, and character actions was sufficient for the anticipated purposes. Generally, the transcripts were designed to offer plenty of contextual information to help set the scene without indicating a subjective interpretation of the communicative events. Figure 1 displays an excerpt from one of the texts used in the corpus, viz. an extract from Six Feet Under, Season 1, Episode 1 (“Pilot”).

Information on the setting is usually given at the beginning of every scene (e.g. line 1). This may include the number of the scene, the location, and the time of day. Character actions are transcribed with varying degrees of detail (e.g. line 2, 4, 7, 22–23, 24). The amount of information primarily depended on the individual fan transcriber of the respective episode and I also added or deleted details whenever appropriate. Hesitation phenomena such as filled pauses, repeats, truncated words, and incomplete utterances were consistently transcribed, as mentioned above (e.g. line 6, 11, 14–15).

3. Television series for a ‘pedagogically relevant corpus’

3.1 What’s so special?

What distinguishes CATS from other corpora is that it uses the dialogues of television series as a data source, which are a) fictional and b) scripted spoken texts. This choice of texts immediately raises a few questions. Why use fictional, i.e. creatively invented speech, rather than ‘real’ speech if the goal is to present students with a natural model of spoken language? [7] A related issue is the validity of using texts which are scripted, i.e. written-to-be-spoken, or in the case of television series, written-to-be-performed. It is important to keep in mind that this corpus uses the transcripts of the actors’ performances, not the scripts by the screenwriters. Nevertheless, the words spoken by the actors are based on a written text. Before I discuss the nature of fictional scripted television language and its validity for classroom use, I want to show that television as a data source for a pedagogical corpus has various advantages over other sources, which is why CATS could be a promising tool for the EFL classroom.

3.2 The medium of ‘television’

For German high school students, television is a popular medium which is not remote from everyday life. The students might even be familiar with the speakers and the dialogues and so they can relate to them, or at least more so than with the rather random and unfamiliar texts in ‘regular’ linguistic corpora. A more practical advantage of television as a data source is that transcripts, which can be used as a first draft, are often freely accessible via the Internet (see above). As imperfect as they may be, they are still a very useful basis, saving the researcher time and resources.

3.3 Contents

The contents represented in CATS correspond to many of the parameters which Braun (2005) lists for a pedagogically relevant corpus. Apart from the students’ familiarity with the genre and possibly the shows per se, the contents are probably also closer to their interests and more relevant to the teaching syllabus. At the center of these shows are everyday topics and interactions between family and friends in their private lives and school/working lives, i.e. the shows cover a wide range of “communicatively relevant topics” (Braun 2006: 30). The four series can be seen as four extended stories, narrated in loosely related chapters, with a limited number of characters/speakers. This makes the corpus texts more coherent and more homogeneous than the random collection of texts found in a regular corpus. In addition, the large amount of information available on the contexts and characters contribute to making a “text-based exploration of the corpus content” (Braun 2006: 29) a conceivable option in the EFL classroom.

3.4 Size

Another factor mentioned by Braun (2005: 49–50) is the size of pedagogically relevant corpora. While mega-sized corpora are extremely valuable for lexicographical research, they are rather difficult for non-linguist users to handle in learning and teaching contexts. Aston (1997: 54) sees many advantages in small corpora for language learning and argues that a size of 20,000 to 200,000 words is probably most suitable for teaching purposes. The appropriate size of a corpus thus ultimately depends on the anticipated use. In the case of CATS, the anticipated use is two-fold: 1.) The corpus should be a tool which allows for the linguistic analysis of various lexico-grammatical characteristics typical of spontaneous spoken language. 2.) The corpus should be a tool which can be used by teachers and learners in an EFL classroom context. Another factor was the limited resources available for the compilation of the corpus. The resulting 160,592 words, however, seem an appropriate basis for both of the anticipated uses.

3.5 Transcription, annotation, data format

As has been mentioned above, much care was taken during the transcription and annotation phase to make the corpus uniform, manageable and easily searchable and thus cater for teacher and learner needs. Another feature of the corpus is that the corresponding audiovisual files can be accessed by means of the official, commercially available DVDs of the television series. [8] These may provide information such as gesture and other non-verbal features of communication. Access to audiovisual data is very useful or sometimes even necessary for an accurate linguistic analysis. For instance, in the case of discourse markers, the decision on whether, say, a you know is a discourse marker or not may depend solely on the intonation because the situational context allows for different readings. Moreover, it is desirable to combine the transcripts with the audiovisual files when teaching spoken grammar to foreign language learners (cf. e.g. Thornbury 2005: 47ff.). Excerpts of the series can be used to demonstrate certain phenomena ‘live’. Learners may also be given transcripts with gaps to fill in while they watch the corresponding excerpt; or they may try to transcribe short parts of the series themselves and then compare them with the ‘original’ transcript. The audiovisual materials furthermore offer plenty of opportunities to connect the more cognitive task of discovering and discussing differences between written and spoken grammar with the practical training of speaking and listening skills or even media-esthetic learning units.

3.6 Language

The nature of the language in CATS is one of the most essential matters to be considered before any conclusions can be drawn on the overall usefulness of CATS for teaching spoken grammar. What is attractive in the EFL context is that television language often represents up-to-date language usage (depending on the exact selection of programs). Contemporary television series can offer the students a window onto current popular language use, such as recent slang terms and other new expressions.

The most interesting and also debatable feature regarding the language in CATS, however, is that it is scripted; it is speech which is based on a written document. For the pedagogical context, it is therefore crucial to explore how this is reflected linguistically, i.e. how exactly television language differs from naturally occurring speech in terms of spoken grammar. How ‘spoken’ is it? Can it be considered natural enough to be a valid model for the EFL classroom? Evaluating the viability of film language as a resource for pragmatic research and language teaching, Rose (2001) comments on this pivotal issue as follows:

[H]ow well does what characters portrayed in film say represent what real-life characters say in face-to-face encounters? If it were to be shown that film language was representative of actual language use, a strong case could be made for its use in the classroom. (Rose 2001: 310)

Of course, matters are more complicated than the question suggests. Rose himself concedes that this is a slightly oversimplified view of the issue, as, for various reasons, there are probably some aspects of the language of film which mirror actual language use quite well while there are other aspects which are represented somewhat differently from naturally occurring speech. The question is whether the differences to be found actually matter. Some might matter to the linguistic description of the nature of television language and at the same time be irrelevant to the EFL teaching context. For instance, a lower frequency of false starts, incomplete utterances and repeats may even be considered beneficial in a classroom context, since it is precisely these features which make work with spoken language and the transcripts complicated for non-linguists. In addition, it is (arguably) not necessary to teach and practice their production anyhow.

Overall, however, it is essential for the grammatical aspects featuring in curricula and syllabi to be represented realistically, not only in terms of quantity (frequency), but also, and this is even more important, in terms of quality (contexts, functions). If it turned out that there was a huge overall discrepancy from naturally occurring speech, i.e. that television language was a ‘gross distortion’, using a television language corpus in class would, of course, not be a good alternative to real spoken corpora when it comes to finding appropriate models of spoken English. The assumption at the beginning of this project is that CATS is a possible in-between solution, featuring a type of language that is less messy and overwhelming than natural spoken language (as present in real spoken corpora), but more natural and motivating than the dialogues in mainstream textbooks.

4. The nature of fictional scripted television language

Investigation into the nature of fictional scripted television language, focusing especially on its potential of acting as a surrogate for naturally occurring speech, is rather scarce. Apart from various unsubstantiated claims and assertions about the nature of television language, very few empirical studies have been conducted. Tatsuki (2006: 6) notes that “there has been virtually no research to assess the validity of film use as an authentic representation of actual language use”. In the same vein, Quaglio (2009: 12) states that “there seems to be a dearth of studies on the language of television from a linguistic point of view”.

The bulk of research has been conducted in the area of pragmatics (e.g. Fernández-Guerra 2008; Martínez-Flor 2008; Rose 2001) and in the field of translation studies (e.g. Mittmann 2006; Taylor 2004, 2008, 2009). There has also been some interest in the area of sociolinguistics (e.g. Tagliamonte and Roberts 2005) and discourse analysis (e.g. Bednarek 2008, 2010). Systematic, large-scale studies on the lexico-grammar of television language are rare, however. An exceptional contribution is the corpus-linguistic study by Quaglio (2008, 2009). His analysis of the situation comedy Friends showed that the language in Friends shares the core linguistic features typical of natural conversation. While he also found that the language in the sitcom is less vague and more emotional and informal than naturally occurring speech, the overall impression remains that television language is strikingly similar to naturally occurring conversation: “[T]he use of television dialogue as a surrogate for natural conversation for the analysis of certain linguistic features seems perfectly appropriate”, especially those that are “less likely to be captured by a corpus of natural conversation” (Quaglio 2009: 149). Quaglio does not specify in the following which exact features these are, but he seems to be referring to linguistic phenomena occurring in settings and situations which are not represented as often in linguistic corpora as they are in television series, e.g. very intimate and highly emotional settings, greetings and leave-takings (see also Quaglio 2009: 111, 115, 148). In other words, some features may be rare in a corpus of natural conversation (compared to television language) not because they are rare in ‘the real world’, but because of the principles of authentic data collection.

One conclusion that seems to find general agreement as regards the lexico-grammar is that television language sits uneasily on the boundary between spoken and written language and it thus comprises a complex mix of spoken and written language features. There are various reasons for this. When spoken language is based on a script, the discourse circumstances are different from natural spontaneous conversation. For instance, the speakers – in this case, the actors who perform – do not face the online planning and production pressure which affects natural conversation taking place in real time (cf. Biber et al. 1999: 1048f.). Almost all parts of the conversation have been planned and prepared for them by the scriptwriters. Spoken language traits which are related to normal planning pressure (e.g. performance phenomena such as hesitations, repeats, and incomplete utterances) are thus unlikely to figure very prominently in scripted language, where speakers know exactly what they are going to say, when they take their turns, how their interlocutor is going to react, etc. Scripted television language could therefore be described as some type of ‘polished material’, as it were.

Another crucial difference between natural spontaneous speech and fictional television speech is the type of addressee. The ultimate addressees of a conversation in a film or television show are the television viewers. The viewers are recipients only, however, and at no point do they have the chance to ask for clarification as in natural communicative circumstances. Everything the characters say to each other has to be acoustically clear and propositionally understandable to the television audience. This means that some spoken features are simply not desirable in television dialogue. Phenomena which are associated with the shared context of the interlocutors and the interactiveness of natural conversation (e.g. vagueness, overlapping speech, incomplete utterances, etc.) are only tolerable to a certain degree. Scriptwriters are responsible for ensuring comprehension by writing scripts that adhere to these requirements, while actors are responsible for performing accordingly. In sum, it has been taken as a given fact that fictional scripted television dialogue diverges in many respects from naturally occurring dialogue, the most striking difference being a lower frequency of performance phenomena and features connected with the interactiveness of natural conversation (cf. Quaglio 2009: 3–4).

As mentioned above, one major purpose of CATS is to contribute to determining a place for television language on the continuum between spoken and written language. The terms ‘written language’ and ‘spoken language’ are in fact quite ambiguous and further comment seems in order. Esser (2006) demarcates the various meanings by distinguishing three aspects of ‘written language’ or ‘spoken language’ (see Table 2). What I am interested in is the “abstract grammatical and lexical form” (Esser 2006: 24) of television language, i.e. to what extent it is characterized by a “written style” or a “spoken style”. To put it in Esser’s (2006) terminology, it is obvious that it is “spoken language” in the sense that the realization of the word forms, the substance, is spoken (i.e. the actors’ performance); it is also clear that it is “written language“ in the sense that it has its origins in writing (i.e. the script); but what is far from certain is how exactly these somewhat ‘mixed’ circumstances are reflected in style.

It is not until we have a better idea of the nature of television language that we can make any claims about its usefulness as a model in EFL teaching. In the following, two features of spoken language are analyzed which can be taken as indicative of ‘spokennness’, i.e. of spoken style. [9]

5. Analysis of two indicators of spoken style

5.1 Filled pauses in CATS

To determine its place on the spoken-written continuum, the first search with CATS was conducted on ‘filled pauses’ (alternatively called ‘fillers’, ‘hesitators’, or, more recently, ‘planners’), namely uh and uhm. [10] These features seemed interesting because they are a prototypical result of online production pressure in natural conversation. Filled pauses buy the speaker time to think, to find the right words and in the meantime hold the floor. However, it is still debated whether and to what extent these items are intentionally applied or simply an involuntary by-product of the rather complex communicative circumstances. [11] The data serving as a reference point for a comparison with British English (BrE) and American English (AmE) conversation come from the Longman Grammar of Spoken and Written English (Biber et al. 1999). Biber et al. (1999: 1096) provide approximate counts in frequency per million words (pmw) for British and American English conversation, which is why this measure was also used for the much smaller CATS corpus. [12] While it is clear that Biber et al.’s numbers are not the ideal reference point for various reasons (e.g. the different composition of corpora, representation of different age groups, variety of topics), it currently appears to be the most appropriate source available for the purposes of the study.

I used WordSmith Tools, Version 4.0 (Scott 2005) to extract all instances of uh and uhm. Figure 2 reveals that uh occurs in CATS at a frequency of almost 3,800 instances pmw, which is markedly less than in naturally occurring AmE conversation (6,500 pmw), but quite comparable to BrE conversation (4,000 pmw). The difference from AmE proves statistically significant in a log-likelihood test with p<0.0001 (LL=207.56), while the small difference from BrE conversation is indeed insignificant (p>0.05; LL=1.8). [13] Considering the ‘scriptedness’ of the dialogues in television series and thus the lack of production pressure and planning difficulties, the number for uh is probably not as low as intuitively expected. Uhm occurs significantly less frequently in CATS than in both corpora of naturally occurring speech (p<0.0001); yet again, considering that we are dealing with a kind of language that was planned long before the production process, an even lower frequency could have been expected. Some possible reasons for the presence of performance phenomena in scripted speech are discussed further below. As a second step, in order to see whether there are any mentionable differences in frequency between the four shows, uh was looked at separately in the four subcorpora. Figure 3 shows how the overall frequency of uh in CATS came about. There is considerable disparity between the four shows: Veronica Mars has around 2,700 instances pmw of uh, which is the lowest frequency, while Monk and Six Feet Under are much closer to naturally occurring AmE conversation (as reported by Biber et al. 1999). In fact, the difference between the subcorpus with the lowest score, Veronica Mars, and the corpus with the highest score, Six Feet Under, is statistically significant with p<0.0001 (LL=19.35). In other words, although we are within one register, i.e. television series or drama series, there is striking variation as regards filled pauses. Finally, it is interesting to note that the frequencies of uh in Monk (4,698 pmw) and in SFU (4,797 pmw) are even higher than in BrE conversation (4,000 pmw). This slight overuse is significant with p<0.05 in both cases (LL=4.37 for Monk and LL=5.45 for SFU).

5.2 Discourse marker well in CATS

The second feature to be analyzed was the discourse marker well. Well is a useful indicator of spoken style because it is a very frequent feature in natural spoken language usage, fulfilling a myriad of structural/textual and interpersonal/interactive functions in conversation (cf. e.g. Aijmer and Simon-Vandenbergen 2003; Cuenca 2008; Müller 2004, 2005). It also appears to be one of the most frequently analyzed discourse markers (cf. Cuenca 2008: 1373). I extracted all instances of well and eliminated those instances which were not discourse markers (e.g. well used as an adverb), following the definition and classification as in Müller (2005). Figure 4 reveals that the frequency of the discourse marker well in CATS with 4,577 instances pmw does not seem to be very far off naturally occurring speech when compared with Biber et al.'s (1999: 1096) data for BrE (5,500 pmw) and AmE (6,000) conversation, even if the differences are significant from a statistical perspective (p<0.0001 in both cases; LL=25.42 for BrE and LL=57.04 for AmE).

CATS is again closer to BrE, but the normalized frequency in CATS resembles that of the Longman AmE conversation subcorpus with a considerable ratio of about 3:4. However, as mentioned before, comparing corpus data with the results of previous analyses of other corpora always bears some risks. The items in question might display quite different frequencies in other corpora, not only because of the different corpus designs (e.g. due to a somewhat different range of settings), but also due to the possibly different ways of defining, classifying, extracting, and analyzing the items, too. For instance, as can be seen in Figure 4, Fung and Carter (2007: 426) report a much lower frequency of well in an alternative BrE speech corpus, viz. the ‘pedagogic sub-corpus’ of the CANCODE (Cambridge and Nottingham Corpus of Discourse in English; see McCarthy 1998: Ch. 1 for further details on the corpus). Their 460,055-word corpus, which represents teacher-student and student-student interactions, contains only about 3,558 instances of well (pmw), considerably fewer than CATS. Conclusions regarding similarity thus always have to be drawn with caution. What is more, keeping the potential pedagogical uses of the corpus in mind, ‘qualitative similarity’ seems to be just as or even more important than purely ‘quantitative similarity’. This is why future analyses will have to consider the representation of the various functions and contexts of well, too (see e.g. Dose in prep.). The next section focuses on the four shows individually.

Figure 5 shows that as in the case of uh, the four television shows in CATS create a very heterogeneous picture when it comes to the discourse marker well. Gilmore Girls basically equals the frequency that Biber et al. (1999) report for their American conversation subcorpus, while Veronica Mars only displays about half of the frequency given for naturally occurring AmE conversation. If we considered only the total numbers for CATS we would miss the fact that some shows resemble naturally occurring speech to a great extent – at least from a purely quantitative perspective – while others display much fewer spoken features. The difference between Gilmore Girls and Veronica Mars in fact proves statistically significant in a log-likelihood test at p<0.0001 (LL=32.77). So, when making claims about the nature of television language, one cannot underestimate the variability among different programs. But why is it that some shows lack features such as filled pauses and discourse markers more than others?

6. Performance phenomena in fictional scripted television dialogue

There are various factors which might help explain the striking variation between the four shows, which have to do with a) the scriptwriting process, b) the actors’ performance, c) the characters in the shows, and d) the types of dialogues represented in each subcorpus. [14]

A common goal of scriptwriting for contemporary television is to represent reality, i.e. the situations, actions, characters, etc. need to appear realistic to the television viewers so that they can identify with the characters and be caught in the illusion that ‘this is really happening’. In general, this also applies to the language used: Scriptwriters will usually try to make the fictional words which they put into the mouths of their characters sound natural (cf. Baumgarten 2003: 20f.; Richardson 2010: 5). After all, it is first and foremost the ‘good’ dialogues which make a show successful, and ‘good’ in this context does not only mean ‘quick-witted’ and ‘clever’ (which serves the viewer’s wish to be entertained), but also ‘natural’ (which serves the viewer’s wish to be caught in the illusion of reality). However, many of the features of naturally occurring speech are beyond awareness and so the frequency of spoken features in a script depends on the scriptwriters’ varying skills of rendering the speech to appear as natural as possible. Richardson (2010: 45) points out some linguists’

suspicion[s] that dramatists themselves don’t fully understand what naturally occurring talk is really like. As ordinary language users, they, like the rest of us, mentally edit out disfluency and other complications in the everyday business of making sense – and carry this deafness over to their representational work. [...] But some dramatists and directors certainly do have an awareness of the ways dialogue can be fashioned that move it away from standard-issue, one-speaker-at-a-time fluency.

Even if scriptwriters are aware of certain features which could be part of a realistic representation of spoken language, we find that the nature of spoken language is a rather neglected area in scriptwriting textbooks and guides. It is highly unlikely that writers will integrate these features into their TV dialogues at the same frequency as they actually occur in ‘real’ speech. In addition, many scriptwriters have a rather negative attitude towards features such as performance phenomena and discourse markers, viewing them as a sign of linguistic deficiency. For example, in his guide on writing for television, Brody (2003: 215) labels what we call ‘discourse markers’ as “like the proverbial plague” and “unnecessary and redundant”. A lack of awareness of the nature and frequency of certain spoken language features may thus be coupled with a marked dislike for them.

The actors who perform the scripted dialogues are the next variables. While it is certainly the job of the scriptwriters to write realistic dialogues, when it comes to items such as filled pauses, discourse markers and hedges, within the film industry it is often seen as the actors’ responsibility, not the scriptwriters’, to add them to the script (cf. Richardson 2010: 65). This applies to performance phenomena in particular: “[S]creenwriting culture constructs its division of labor between writers and actors through a practical implementation of the general folk-linguistic understanding that expressions of disfluency are not part of the (verbal) meaning but instead are performance errors” (Richardson 2010: 65, emphasis in the original). In consequence, the extent to which performance phenomena such as filled pauses are represented in the dialogues as they are finally performed and broadcast depends on the different degrees to which actors spontaneously prime in spoken language features while performing. By adding these features to the script, actors try to make the dialogue ‘less written’ and ‘more spoken’, although they might be restricted by the scriptwriters’ instructions, who often do not permit major changes to their ‘creative work’ (cf. Wray 2008: 183). Yet, in his investigation of discourse markers such as well, you know, you see, and filled pauses such as um in the feature film Notting Hill (1998), Taylor (2004) found that there were great differences in frequency between the original film script and the transcripts of the words as they were actually performed by the actors. [15] [16] The spoken language phenomena featured much more frequently in the transcriptions than in the film scripts. Taylor concludes that

[i]f all the oft identified features of spoken language (hesitation, repetition, ellipsis of subject pronouns, auxiliaries, articles and initial parts of set expressions, pre-and post-placed items, etc.) are ‘primed out’ in scripts, it seems that they are to some extent primed in again by the actors when they interact. (Taylor 2004: 80)

Of course, the extent of such ‘priming in’ varies from actor to actor and so it can have an impact on the frequency of spoken language features in different television shows. Apart from the general skills, preferences, and idiosyncrasies displayed by scriptwriters and actors, there is an additional factor influencing language use, namely characterization on television.

Above, I have considered character-independent use of spoken language features by scriptwriters and actors. However, especially in the case of filled pauses, the degree to which these are used might actually be an integral part of the character that is being performed. In other words, both scriptwriter and actor might intentionally put an exceptional number of fillers in a character’s mouth as a way of characterizing, of emphasizing his or her personal traits. For example, in the case of the show Monk, the protagonist is a very hesitant and insecure person by nature and so it is possibly for this reason that Monk displays a high frequency of uh. The type of dialogue structure which is dominant in a given television show may affect the frequency of certain spoken features, too. Gilmore Girls, for instance, is well-known for its very fast-paced dialogues, implying not only a higher articulation rate and more spoken words per episode, but also more frequent turn-taking than in other shows. Since the discourse marker well often occurs in the beginning of a turn (cf. e.g. Biber et al. 1999: 1086; Aijmer 2011: 234f.), it is not surprising, if such a show displays a remarkably high frequency of well, as there are simply more slots for well to occur. Veronica Mars, in turn, is somewhat of an ‘odd one out’ among the shows used in CATS because it is the only show in this corpus to include some extended stretches of non-dialogic speech (e.g. voice-overs), which were intentionally not removed from the transcript during corpus compilation due to pedagogical considerations. This might influence the results in that there may be fewer slots in the sample for items such as discourse markers from the start. Further qualitative analyses need to be conducted in order to substantiate these assumptions.

7. Summary and conclusion

In this paper, I first described and discussed the design and compilation of a new resource for studying fictional scripted television language and for teaching and learning spoken English grammar: CATS, the Corpus of American Television Series. The corpus reflects the necessary characteristics that have been posited for a ‘pedagogically relevant corpus’ (Braun 2005) to a large extent, allowing for exploitation from various angles which comply with the local syllabi and curricula.

The central question is, however, whether the language as represented in television series is an appropriate model for language learners, since it is unclear how scripted television language differs from naturally occurring language and whether these differences actually matter. While television language is generally an under-researched field, it is certain that it is a variety which is situated in a fuzzy area between speech and writing, between spokenness and writtenness (due to e.g. the different communicative circumstances). CATS will be a new resource to help explore this matter. As a first step, two features which were deemed to be good ‘indicators of spoken style’ were chosen for a quantitative analysis: Filled pauses (uh, uhm) and the discourse marker well. The results for filled pauses indicated that television language is indeed less ‘messy’ than naturally occurring language (i.e. displaying a lower frequency of filled pauses), but it is by no means entirely free of performance phenomena. Actors consciously or unconsciously add spoken language features to the script to make the dialogues sound more natural. With regard to these performance phenomena, then, fictional television language might prove to be the perfect middle-ground between real-language material such as linguistic corpora of spontaneous spoken language (the latter displaying an abundance of performance phenomena unwelcome in the teaching context) and the often rather artificial, stiff and written-like representations of spoken language in traditional textbooks.

Nevertheless, it was also shown that there are quite striking differences between the four shows with respect to the frequency of filled pauses. Some shows (in this case, Monk and Six Feet Under) are much closer to natural spoken style than others. At the same time, the varying frequencies also depend on the characters of the shows, just as they depend on individuals in the real world. The second feature which was analyzed, the discourse marker well, seems to be fairly well represented from a frequency-focused point of view with a ratio of about 3:4 compared to AmE conversation (cf. Biber et al. 1999). A separate look at the individual shows revealed that different features are represented to varying extents in the four shows. For instance, the series Gilmore Girls turned out to have the same frequency of well as reported by Biber et al. (1999) for naturally occurring spoken language (AmE). Six Feet Under scored quite high as well, while Monk and Veronica Mars lagged far behind and thus caused the total frequency of CATS to be lowered. The conclusion we might draw from these individual numbers is that fictional scripted television language is too diverse and heterogeneous, too dependent on the characters, individual writing styles of scriptwriters, and performance styles of actors to ‘lump everything together’ and make generalized statements about television language and its potential to act as a surrogate for natural speech. In consequence, judgments about the appropriateness of fictional scripted television language for teaching spoken grammar will probably also have to be more discriminating and take account of the variation between different shows.

The items analyzed for the present paper were only the first exploratory steps of a larger, systematic analysis (Dose in prep.). One major aim will be to assess the status of fictional scripted television language as represented in CATS on the scale between spoken and written style. To this end, an analysis of a variety of performance phenomena and other spoken language features will be conducted from a quantitative and qualitative perspective. Special emphasis, particularly as regards typical functions and contexts, will then be placed on features which are considered relevant to the teaching context. For instance, while features such as hesitation and repeats help a great deal in assessing the level of spoken style, it is improbable that they would or should need particular attention in the classroom; it is enough to make the students aware of them. In turn, features such as discourse markers are extremely useful for improving oral competence. Since one major aim of the project is to establish whether or to what extent CATS can provide an appropriate model of spoken English to EFL learners, the pedagogically relevant features will be the focus in the qualitative analyses.

I hope to have shown in the present paper that CATS is a useful new resource for the description of fictional scripted television language. It is hoped that the results of future analyses will prove CATS to be a useful resource for teaching spoken grammar in the EFL classroom, too.

Notes

[1] I would like to thank Rosemary Bock for proofreading the original manuscript. All remaining errors are, of course, solely my responsibility.

[2] Note that the present paper and Dose (2012) are based on a preliminary version (2010 version) of CATS. The final version of CATS is used and described in more detail in Dose (in prep.).

[3] The years the shows ran in Germany differ slightly, usually lagging a couple of years behind the US market. All shows have aired their final episodes by now and will not be renewed, but the old episodes regularly re-run on American and German television.

[4] Unfortunately, the website www.twiztv.com has gone down since the compilation of CATS. The Internet Archive Wayback Machine offers a number of archived snapshots of this website, however. A snapshot of 30 Aug. 2010 can be accessed at http://web.archive.org/web/20100830040325/http:/www.twiztv.com/ (last accessed 11 April 2013).

[5] While an orthographic transcription can never capture all the manifold phenomena of spoken language, it is usually deemed sufficient for simple lexical and grammatical corpus analyses (cf. Leech 2000: 678).

[6] The detailed documentation of performance phenomena distinguishes CATS from other pedagogical corpora (see e.g. Braun 2007a: 35 for details on the transcription practice in ELISA).

[7] A consideration of the ‘authenticity debate’ in EFL teaching is beyond the scope of the present article. Suffice it to say that the view that naturally occurring language should be preferred over specially invented language (such as constructed textbook dialogues) as a model for learners – the standpoint of many researchers in a corpus-linguistic tradition, e.g. Aston (2000) – is not shared by all linguists and foreign language teaching experts (see e.g. Cook 2001 and Widdowson 1998 for a discussion of this issue).

[8] The audiovisual files may be aligned with the transcripts at a later stage of this project.

[9] The choice of features to be analyzed here was somewhat arbitrary and only marks the beginning of a larger, systematic analysis of a variety of spoken language features (Dose in prep.).

[10] Note that filled pauses are commonly transcribed as uh and uhm in American English and as er and erm in British English. However, this does not imply a different pronunciation, i.e. uh and er can be considered the same item (cf. Biber et al. 1999: 1053). Since CATS strictly follows American English spelling conventions, filled pauses were consistently transcribed as uh and uhm.

[11] For a discussion of the functions and meanings of uh and uhm, see e.g. Clark and Fox (2002), Corley and Stewart (2008), and Kjellmer (2003). The perspective taken on this matter does not only influence the choice of terminology (e.g. ‘hesitators’ vs. ‘planners’), but also has consequences for the pedagogical context. If fillers such as uh and uhm are considered to be used by native speakers as some sort of ‘communication strategy’ (cf. e.g. Dörnyei 1995: 57–59), there would be a case for teaching their production in the EFL classroom. In contrast, if they are viewed as an involuntary side effect of online production pressure, they may not deserve particular attention beyond raising awareness of their existence.

[12] While it is generally preferable to use a base near the size of the smallest corpus used in the comparison, I nonetheless opted to norm all frequencies to a base of 1,000,000 (instead of, say, 100,000) because a) this was the base used in the reference data for natural spoken language (Biber et al. 1999) and b) this makes it easier to compare results with other studies, which tend to report results in occurrences pmw.

[13] All log-likelihood values were calculated with Paul Rayson's online log-likelihood calculator (http://ucrel.lancs.ac.uk/llwizard.html, 30 Nov. 2011). Since Biber et al. (1999) do not provide raw frequencies to be used in the log-likelihood tests, these were determined based on Biber et al.'s normalized counts and the respective corpus sizes.

[14] A more elaborate framework which captures the manifold factors influencing the degree of spoken style in fictional scripted television dialogue is offered in Dose (in prep.).

[15] The um in Taylor’s (2004) study is a variant transcription equivalent to uhm and erm.

[16] The fact that there are substantial changes between the original script and the actual acting out of the script (see also Mittmann 2006: 579 and Wray 2008: 180–182 on this issue) means that a language researcher analyzing the nature of television language should use the transcripts, not the scripts of the corresponding shows as a data base.

Sources

twiztv.com: www.twiztv.com. The site is no longer online, but a cached version of it can be seen via the Internet Archive Wayback Machine: http://web.archive.org/web/20100830040325/http:/www.twiztv.com/

Comprehensive Episode Guides: http://episodeguides.blogspot.com/

References

Aijmer, Karin. 2011. “Well I’m not sure I think... The use of well by non-native speakers”. International Journal of Corpus Linguistics 16(2): 231–254.

Aijmer, Karin & Anne-Marie Simon-Vandenbergen. 2003. “The discourse particle well and its equivalents in Swedish and Dutch”. Linguistics 41(6): 1123–1161.

Aston, Guy. 1997. “Small and large corpora in language learning”. Proceedings of PALC 1997, ed. by Barbara Lewandowska-Tomaszczyk & Patrick James Melia, 51–62. Lodz: Lodz University Press.

Aston, Guy. 2000. “Corpora and language teaching”. Rethinking Language Pedagogy from a Corpus Perspective: Papers from the Third International Conference on Teaching and Language Corpora, ed. by Lou Burnard & Tony McEnery, 7–17. Frankfurt am Main: Peter Lang.

Baumgarten, Nicole. 2003. “Shaken and stirred: Language in film in a cross-cultural perspective”. Zeitschrift für Interkulturellen Fremdsprachenunterricht 8(2/3): 20–29.

Bednarek, Monika. 2008. “‘What the hell is wrong with you?’ A corpus perspective on evaluation and emotion in contemporary American pop culture”. Questioning Linguistics, ed. by Ahmar Mahboob & Naomi Knight, 95–126. Newcastle: Cambridge Scholars Press.

Bednarek, Monika. 2010. The Language of Fictional Television: Drama and Identity. London: Continuum.

Biber, Douglas, Stig Johansson, Geoffrey Leech, Susan Conrad & Edward Finegan. 1999. Longman Grammar of Spoken and Written English. Harlow: Pearson Education.

Braun, Sabine, ed. 2004. ELISA English Language Interview Corpus as a Second-Language Application http://www.uni-tuebingen.de/elisa [archive.org].

Braun, Sabine. 2005. “From pedagogically relevant corpora to authentic language learning contents”. ReCALL 17(1): 47–64.

Braun, Sabine. 2006. “ELISA: A pedagogically enriched corpus for language learning purposes”. Corpus Technology and Language Pedagogy: New Resources, New Tools, New Methods, ed. by Sabine Braun, Kurt Kohn & Joybrato Mukherjee, 25–47. Frankfurt am Main: Peter Lang.

Braun, Sabine. 2007a. “Designing and exploiting small multimedia corpora for autonomous learning and teaching”. Corpora in the Foreign Language Classroom: Selected Papers from the Sixth International Conference on Teaching and Language Corpora (TaLC 6), University of Granada, Spain, 4–7 July 2004, ed. by Encarnación Hidalgo, Luis Quereda & Juan Santana, 31–46. Amsterdam: Rodopi.

Braun, Sabine. 2007b. “Integrating corpus work into secondary education: From data-driven learning to needs-driven corpora”. ReCALL 19(3): 307–328.

Braun, Sabine & Angela Chambers. 2006. “Elektronische Korpora als Ressource für den Fremdsprachenunterricht”. Praktische Handreichung für Fremdsprachenlehrer (4th ed.), ed. by Udo O.H. Jung, 330–337. Frankfurt am Main: Peter Lang.

Brody, Larry. 2003. Television Writing from the Inside Out: Your Channel to Success. New York: Applause Cinema and Theatre Books.

Carter, Ronald & Michael McCarthy. 2006. Cambridge Grammar of English: A Comprehensive Guide. Spoken and Written English Grammar and Usage. Cambridge: Cambridge University Press.

Clark, Herbert H. & Jean E. Fox Tree. 2002. “Using uh and um in spontaneous speaking”. Cognition 84(1): 73–111.

Cook, Guy. 2001. “‘The philosopher pulled the lower jaw of the hen’. Ludicrous invented sentences in language teaching”. Applied Linguistics 22(3): 366–387.

Corley, Martin & Oliver W. Stewart. 2008. “Hesitation disfluencies in spontaneous speech: The meaning of um”. Language and Linguistics Compass 2(4): 589–602.

Council of Europe. 2001. Common European Framework of Reference for Languages: Learning, Teaching, Assessment. Cambridge: Cambridge University Press.

Cuenca, Maria-Josep. 2008. “Pragmatic markers in contrast: The case of well”. Journal of Pragmatics 40(8): 1373–1391.

Dörnyei, Zoltán. 1995. “On the teachability of communication strategies”. TESOL Quarterly 29(1): 55–84.

Dose, Stefanie. 2012. "Scripted speech in the EFL classroom: The Corpus of American Television Series for teaching spoken English". Input, Process and Product: Developments in Teaching and Language Corpora, ed. by James Thomas & Alex Boulton, 62–80. Brno: Masaryk University Press.

Dose, Stefanie. in prep. Describing and Teaching Spoken English: An Educational-Linguistic Study of Scripted Speech.

Esser, Jürgen. 2006. Presentation in Language: Rethinking Speech and Writing. Tübingen: Narr.

Fernández-Guerra, Ana. 2008. “Requests in TV series and in naturally occurring discourse: A comparison”. Learning How to Request in an Instructed Language Learning Context, ed. by Eva Alcón-Soler, 111–126. Frankfurt am Main: Peter Lang.

Fung, Loretta & Ronald Carter. 2007. “Discourse markers and spoken English: Native and learner use in pedagogic settings”. Applied Linguistics 28(3): 410–439.

Hughes, Rebecca. 2002. Teaching and Researching Speaking. Harlow: Longman.

Kieweg, Werner. 2000. “Zur Mündlichkeit im Englischunterricht”. Der Fremdsprachliche Unterricht Englisch 47: 4–9.

Kjellmer, Göran. 2003. “Hesitation: In defence of ER and ERM”. English Studies 84(2): 170–198.

Kurtz, Jürgen. 2001. Improvisierendes Sprechen im Fremdsprachenunterricht: Eine Untersuchung zur Entwicklung spontansprachlicher Handlungskompetenz in der Zielsprache. Tübingen: Narr.

Leech, Geoffrey. 2000. “Grammars of spoken English: New outcomes of corpus-oriented research”. Language Learning 50(4): 675–724.

Martínez-Flor, Alicia. 2008. “Analysing request modification devices in films: Implications for pragmatic learning in instructed foreign language contexts”. Intercultural Language Use and Language Learning, ed. by Eva Alcón-Soler & Maria Pilar Safont Jordà, 245–280. Dordrecht: Springer Science+Business Media B.V.

Mauranen, Anna. 2004. “Speech corpora in the classroom”. Corpora and Language Learners, ed. by Guy Aston, Silvia Bernardini & Dominic Stewart, 195–211. Amsterdam: John Benjamins.

McCarthy, Michael. 1998. Spoken Language and Applied Linguistics. Cambridge: Cambridge University Press.

Meunier, Fanny. 2002. “The pedagogical value of native and learner corpora in EFL grammar teaching”. Computer Learner Corpora, Second Language Acquisition, and Foreign Language Teaching, ed. by Sylviane Granger, Joseph Hung & Stéphanie Petch-Tyson, 119–141. Amsterdam: John Benjamins.

Mittmann, Brigitta. 2006. “With a little help from Friends (and others): Lexico-pragmatic characteristics of original and dubbed film dialogue”. Anglistentag 2005 Bamberg: Proceedings of the Conference of the German Association of University Teachers of English, ed. by Christoph Houswitschka, Gabriele Knappe & Anja Müller, 573–585. Trier: Wissenschaftlicher Verlag Trier.

Müller, Simone. 2004. “‘Well you know that type of person’: Functions of well in the speech of American and German students”. Journal of Pragmatics 36(6): 1157–1182.

Müller, Simone. 2005. Discourse Markers in Native and Non-Native English Discourse. Amsterdam: John Benjamins.

Mukherjee, Joybrato. 2004. “Bridging the gap between applied corpus linguistics and the reality of English language teaching in Germany”. Applied Corpus Linguistics: A Multidimensional Perspective, ed. by Ulla Connor & Thomas Upton, 239–250. Amsterdam: Rodopi.

Mukherjee, Joybrato. 2009. “The grammar of conversation in advanced spoken learner English: Learner corpus data and language-pedagogical implications”. Corpora and Language Teaching, ed. by Karin Aijmer, 203–230. Amsterdam: John Benjamins.

Quaglio, Paulo. 2008. “Television dialogue and natural conversation: Linguistic similarities and functional differences”. Corpora and Discourse: The Challenges of Different Settings, ed. by Annelie Ädel & Randi Reppen, 189–209. Amsterdam: John Benjamins.

Quaglio, Paulo. 2009. Television Dialogue: The Sitcom Friends vs. Natural Conversation. Amsterdam: John Benjamins.

Richardson, Kay. 2010. Television Dramatic Dialogue: A Sociolinguistic Study. New York: Oxford University Press.

Römer, Ute. 2008. “Corpora and language teaching”. Corpus Linguistics: An International Handbook, ed. by Anke Lüdeling & Merja Kytö, 112–131. Berlin: de Gruyter.

Rose, Kenneth R. 2001. “Compliments and compliment responses in film: Implications for pragmatics research and language teaching”. International Review of Applied Linguistics 39(4): 309–336.

Scott, Mike. 2005. WordSmith Tools (4.0). Oxford: Oxford University Press.

Tagliamonte, Sali & Chris Roberts. 2005. “So weird; so cool; so innovative: The use of intensifiers in the television series Friends”. American Speech 80(3): 280–300.

Tatsuki, Donna. 2006. “What is authenticity?”. Authentic Communication: Proceedings of the 5th Annual JALT Pan-SIG Conference, May 13–14, 2006, ed. by Tim Newfields, Ian Gledall, Megumi Kawate-Mierzejewska, Yvonne Ishida, Mark Chapman & Peter Ross, 1–15. Shizuoka, Japan: Tokai University College of Marine Science. http://jalt.org/pansig/2006/HTML/Tatsuki.htm

Taubenböck, Andrea. 2007. “Sprache kommt von Sprechen: Ein Plädoyer für mehr Mündlichkeit im Englischunterricht”. Der Fremdsprachliche Unterricht Englisch 90: 2–9.

Taylor, Christopher. 2004. “The language of film: Corpora and statistics in the search for authenticity. Notting Hill (1998) – a case study”. Miscelánea: A Journal of English and American Studies 30: 71–85.

Taylor, Christopher. 2008. “Predictability in film language: Corpus-assisted research”. Corpora for University Language Teachers, ed. by Carol Taylor Torsello, Katherine Ackerley & Erik Castello, 167–181. Frankfurt am Main: Peter Lang.

Taylor, Christopher. 2009. “‘I knew he'd say that!’ A consideration of the predictability of language use in film”. Translation: Neue Entwicklungen in Theorie und Praxis: SummerTrans-Lektionen zur Translationswissenschaft, ed. by Lew N. Zybatow, 173–186. Frankfurt am Main: Peter Lang.

Thornbury, Scott. 2005. How to Teach Speaking. Harlow: Longman.

Timmis, Ivor. 2005. “Towards a framework for teaching spoken grammar”. ELT Journal 59(2): 117–125.

Widdowson, Henry G. 1998. “Context, community, and authentic language”. TESOL Quarterly 32(4): 705–716.

Wray, Alison. 2008. Formulaic Language: Pushing the Boundaries. Oxford: Oxford University Press.

Zorzi, Daniela. 2001. “The pedagogic use of spoken corpora: Introducing corpus concordancing in the classroom”. Learning with Corpora, ed. by Guy Aston, 85–107. Houston, Texas: Athelstan.