Corpora in the classroom: Data-driven learning for Freshman English

Simon Smith

Foreign Language Center, National Chengchi University

[email protected]
Abstract
In recent years, Western ELT scholarship has emphasized student-centered learning,
focusing in particular on Computer-Aided Language Learning (CALL) and the use of
linguistic corpora, including Data-Driven Learning (DDL). In the universities of Taiwan, and
many Asian nations, those teaching English to non-majors have to overcome certain hurdles
if they are to successfully use DDL techniques. Class numbers tend to be high, with up to 70
students in the room, and computer workstations for individual use are not normally available.
The present paper, therefore, shows how to make selective use of salient corpus examples
before the whole class, for example when explaining language points in warm-up or wrap-up
sessions. We focus in particular on the importance of making students aware of collocational
patterns.
DDL tasks are often thought of as part of a student-centered research approach, and
some experienced teachers believe that many non-major students lack the motivation or
proficiency to engage with such tasks. The present paper compares that view to that of some
European and Korean studies, where it is claimed that students at even low proficiency levels
can benefit from DDL.
In an approach partly inspired by Content-Based Instruction, we try to motivate our
students to learn from corpora by assigning real-world tasks. The tasks vary in nature, but
often require the students to comment on a software or corpus usability issue, or to use (or
develop) their own computer skills to present vocabulary or other corpus information in an
effective way.
Introduction and background literature
It comes as no surprise, in this digital age, that many approaches to language teaching
and learning rely on the use of the computer, and that countless computer-based tools for
language learning have been developed over the last few decades.
Computer-assisted learning (CALL) is now of great significance and importance in the
acquisition of second languages especially English in Taiwan and all over the world.
There are entire journals devoted to research on the topic, including for example Computer
Assisted Language Learning, published by Taylor and Francis. At nearly all language
teaching institutions, the use of online resources is now commonplace in the language
classroom, and listening labs are often computerized. Ellis (1995) notes that CALL has a
particularly important role to play in the acquisition of vocabulary, because this is the part of
language study to which the student can most usefully turn his attention in private. Thus,
teacher contact hours can be devoted to more communicative activities that cannot so easily
be practised alone. There is, indeed, a great variety of applications available on the web for
students to use in private study. Many ESL textbooks now have accompanying websites with
interactive exercises, and there are CALL sites targeting grammar, such as WERTi (Metcalf
and Meurers 2006; http://prospero.ling.ohio-state.edu/WERTi). Lextutor
(http://www.lextutor.ca) offers vocabulary building exercises, while VISL (Bick 2005;
http://beta.visl.sdu.dk/visl2) features both lexical and grammar-based tasks and games,
including one where the learner is asked to shoot the adjective!
Our own TEDDCloG (Smith et al, forthcoming) system teaches and tests collocational
knowledge through cloze exercises, and is based on the Sketch Engine suite of corpus query
tools.
The use of linguistic corpora in language learning is often described in terms of data
driven learning (DDL), described by Hadley (2002). In a clear parallel to data-driven
computational algorithms, DDL attempts to impart linguistic knowledge by making available
samples of authentic language, and inviting language learners to tease out the grammatical
patterns for themselves. Other, more traditional, approaches teach the grammatical patterns
explicitly, and make use of non-authentic materials, on the analogy of rules-based algorithms.
Johns (1991) likens the language learner (on the DDL model) as a researcher, analyzing
target language data and becoming familiar with the language through the regularities and
consistencies encountered. He notes that it is intelligent, sophisticated, and well-motivated
students that are in the best position to benefit from DDL.
Hadley (2002) mentions resistance to the use of DDL in an Asian context, and notes his
own initial uncertainty about using the approach in his Japanese ELT classroom, on the
grounds that some Japanese are uncomfortable about being taught grammar, especially by a
non-Japanese teacher. This concern probably applies to the Taiwan context also: university
students here sometimes feel that they have already in some sense acquired English grammar,
and all that remains for the perfection of their knowledge of English is to learn more
vocabulary. As many practising teachers at university level will testify, this is in fact very far
from the truth. Furthermore, DDL techniques can be just as effectively applied to the
acquisition of vocabulary as to that of grammar.
Regrettably, the way that target language vocabulary is generally acquired in Taiwan
flies in the face of modern pedagogy. A highly traditional approach is normally taken, where
the student memorizes items from a list: this is probably the only learner-centred task in
which the average university student engages. No special merit seems to be attached to wide
reading or inferring of meaning from context, such as might lead to long-term retention. It is
commonly assumed that almost every word of English has an exact Chinese equivalent to
which its meaning may be mapped; the perceived goal is to figure out which words are likely
to come up in a future test and memorize these mappings.
Another potential objection is that learning through introspection and reflection does not
seem to fit too comfortably into the received model of Asian pedagogy. Most stakeholders
(students, parents, university administrators, education officials and unfortunately many
teachers) have a fairly traditional understanding of the learning process, where essentially the
teacher delivers content and the students somehow absorb it (or the students practise
production, and the teacher corrects them where necessary). This is despite the best efforts of
Taiwan universities, both private and public, to encourage and reward student-centred
learning.
The problem is exacerbated in situations where class sizes are large, and students are
relatively unmotivated. Even if it could be readily demonstrated that DDL yields better
acquisition results than traditional approaches (that is, DDL-taught students are more likely to
acquire written and spoken fluency in the target language), this would not necessarily prove
that the needs of Taiwanese university students were being met. Communicative competence
is only one of a range of such needs; others can include passing proficiency tests (which may
only make a partial assessment of such competence), meeting a stipulated number of teacher
contact hours, socializing with classmates from other academic departments, and learning
English vocabulary for use in Mandarin-English code-switching exchanges, quite common in
Taiwanese professional life. If the teacher is not Taiwanese, cross-cultural communication
needs may be addressed, and skills acquired.
Despite all this, Hadley (2002) and Lee (2007, a Korean high school study) reported
some success with the approach in Asian classrooms. Also, Boulton (2007) described an
experiment in which relatively unmotivated French students reacted positively to a DDL
treatment of English phrasal verb instruction. The lack of motivation that Boulton identifies is
not dissimilar to that often encountered in Taiwan; namely, that the English course being
taken is to fulfil credit requirements only, and the student is unlikely to need English skills in
future professional life.
Boultons experiment consisted of three phases. First, there was a pre-test, where
students ability to choose between pick or pick up, and look or look up was examined. Next,
the students were provided with a concordance sample of 25 lines featuring the same four
lexical items. They were given 10 minutes to peruse the data. Finally, a post-test similar to
the pre-test was administered, and it was found that the performance of all levels of learner
did indeed improve. Boulton is very cautious about his findings, and does not conclude from
them that DDL is successful as an approach. Rather, he hopes that other teachers will conduct
related experiments with their own students, and that ultimately the research community as a
whole will reach firmer conclusions about the usefulness of DDL.
Another point that needs to be made about Boultons work is that the whole procedure
took only half an hour of class time; and that of that, only 10 minutes was spent on reading
the concordance. Poring over this rather dry data for a short time might be acceptable for
most students, but any more than that and boredom is likely to set in, with affective filters
locking securely up in all but the most motivated students. As Kilgarriff et al (2008) put it,
The bald fact is that reading concordances is too tough for most learners. Reading
concordances is an advanced linguistic skill.
The Sketch Engine
Kilgarriffs Sketch Engine (SkE) corpus query toolset goes some way towards
softening concordance data up for language learners. Concordances themselves are
enhanced by making available a sentence mode, as well as the traditional KWIC mode, so
that more may be gathered from the context. Concordance lines can also be ranked by quality:
a good example sentence is defined by Kilgarriff et al as one which is neither too short nor
too long, which doesnt contain a lot of rare words or anaphors (which can sometimes only be
resolved by looking outside the sentence), and is constrained by a few other parameters
specified by the team. The first page of concordance sentences for the word corpus is shown
in Figure 1, in default order, while Figure 2 shows the best examples first. The results
speak for themselves.

Figure 1 Concordance output for corpus, default order

Figure 2 Concordance output for corpus, best-first order

 SkE offers a number of other features which were devised principally with the
lexicographer in mind, but which nevertheless could make a language learners DDL
experience richer and more fun than reliance on raw corpora. These include Word Sketches,
one-page, automatic, corpus-derived summaries of a words grammatical and collocational
behaviour; a distributional thesaurus, which shows which words commonly occur in the same
context as a user-supplied keyword, and are likely to be near synonyms, and the sketch
differences module, which shows how the collocational behaviour of two user-supplied
keywords differs. All of these were described in some detail at the DAE conference two years
ago, in Smith, Huang, Kilgarriff, & Chen (2007).
Corpora in a Taiwan classroom
Over the last two years, the author has been using corpora in Freshman English classes
at two Taiwanese universities, one a private institution, the other public. On the whole,
standards of motivation and attainment are expected to be higher at a public university, and
that expectation has been borne out by impressionistic results. At the private university, I
only ever used corpora in the classroom itself, finding that the meaning and collocational
patterns of a word could be quite well illustrated by projecting a Word Sketch, for example,
on the in-class screen.
In a similar way, I often project Google image search results, and definitions from
online dictionaries, on the screen. I dont think of Word Sketch display as any closer to DDL
than these two procedures, however.
Even though response times on the SkE server are consistently quite fast, it does seem to
take time to call up the displays, and it interferes with the pacing of the class. I also found it
difficult to figure out the right amount of exposure to corpora for students: if too little, they
will perhaps not be familiar enough with the corpus tools to learn much from the displays. If
too much, there is a risk that the lesson will turn into a computer walkthrough and some
students will lose interest.
At the public university, I have continued to make limited in-class use of Sketch Engine
output. I have also given some homework assignments which required students to make use
of corpora, asking them to produce word webs similar to those suggested in the reading skills
textbook I use, Anderson (2008). Among the assignment types I have used, it is this that most
closely resembles a DDL task.

Figure 3 Word web from Anderson (2008)

 As seen from Figures 4 and 5, where students had to choose keywords from a BBC news
story of theirI modified the Anderson format, such that synonyms and antonyms were not
required, but a translation was, as well as example sentences from the concordance. I hoped
that students would find highly salient collocations from Sketch Engine word sketches, and
then click on the concordance function to find sentences which included those collocations.
Figure 4 was turned in by student who understood that part of the task, while Figure 5
contains example sentences which do not illustrate the collocations given.

1. ;;

2. ,
compulsive
education

insurance compulsion

order Collocation Compulsory Word Family

s

admission compulsively
competitive
Examples compulsorily

1. Lessons about personal, social and health matters including sex and
relationships will be compulsory in all England's schools from ages five to 16.
2. Education is compulsory for children in most countries.
3. Whilst contents insurance isn't compulsory, it is strongly advisable.
4. First, ASWs took responsibility for decisions diverting individuals from
compulsory admission.

Figure 4 Student word web demonstrating understanding of collocation examples

Figure 5 Student word web with unrelated collocations and example sentences
In an earlier task, the students had been asked to draw similar word webs by hand and
hand them in on paper; there was great variance among the students artistic skills. For this
second task, an important feature was in fact the production of an acceptable computer
graphic. Students were given no guidance on the file format to use, and had only the example
from the textbook (Figure 3) as inspiration for their choice of layout. Thus, they were
required to rely on their own initiative to complete the task. The task also afforded some
training in computer text and graphics manipulation, which will serve them well in future
university and professional life.
The second semester: future plans for DDL
In the upcoming semester, I shall be devoting more class time to DDL and corpus
linguistics: about one out of two hours total contact time per week. At least two of the
weekly classes will take the form of workshops, where each student will have a computer
available, and they will be able to practise generating concordances and using Sketch Engine
and other corpus query tools under fairly close supervision. At the first workshop, students
will complete a task sheet which I have already used successfully with linguistics
postgraduates; by the end of the semester, they will have to turn in a corpus-based project,
either discussing language patterns that they find interesting, or comparing two or more
corpora or corpus interfaces.
The reader may note some similarity between my planned approach and Content-based
Instruction (CBI), defined Crandall and Tucker (1990:187) as an approach to language
instruction that integrates the presentation of topics or tasks from subject matter classes (e.g.,
math, social studies) within the context of teaching a second or foreign language. In this
case, the subject matter would be corpus linguistics. On another definition (Davies 2003),
CBI means learning about something rather than learning about language. Clearly, learning
about language is one of the goals of the plan. It has been argued by the above scholars and
many others that learners are more likely to acquire language if the subject matter is of
genuine interest to them; I believe that while Taiwanese learners are not drawn with
enthusiasm to the memorization of vocabulary and grammatical patterns, but that they are
interested in language itself, and will enjoy learning about words, collocation and meaning
through the analysis of the authentic texts found in a corpus.
Bibliography
Anderson, N. J. 2008. Active Skills for Reading: Book 4. Boston:Heinle
Bick, E. 2005. Grammar for Fun: IT-based Grammar Learning with VISL. In Henriksen, P. J.
(ed.), CALL for the Nordic Languages. pp.49-64. Copenhagen: Samfundslitteratur
Boulton, A. 2007 Looking (for) empirical evidence of DDL at lower levels. Proceedings of
Practical Applications in Language and Computers 2007. Lodz, Poland.
Crandall, J. & Tucker, G.R. 1990. Content-Based Instruction in Second and Foreign
Languages. In Padilla, A., Fairchild, H.H. & Valadez, C. (eds.) Foreign Language
Education: Issues and Strategies. Newbury Park, CA: Sage.
Davies, S. 2003. Content Based Instruction in EFL Contexts. Internet TESL Journal, Vol. IX,
No. 2, at http://iteslj.org/Articles/Davies-CBI.html
Ellis, R. 1995. Modified oral input and the acquisition of word meanings. Applied Linguistics
16/4, pp. 409-441.
Hadley, G. 2002. Sensing the Winds of Change: An Introduction to Data-Driven Learning.
RELC Journal 33 (2), pp. 99-124.
Johns, T.F. 1991 Should you be persuaded: Two examples of data-driven learning. In Johns,
T.F. and King, P. (Eds.) Classroom Concordancing. (pp. 1-13) Birmingham: ELR
Kilgarriff, A., Husak, M., McAdam, K., Rundell, M. and Rychl, P. 2008. GDEX:
Automatically finding good dictionary examples in a corpus. Proceedings of
EURALEX. Barcelona.
Lee, D. J. 2007 Exploring corpora for Korean secondary school EFL learning: A computer-
aided error analysis and data-driven learning. Proceedings of Practical Applications in
Language and Computers 2007. Lodz, Poland.
Smith, S., Huang, C., Kilgarriff, A., Chen, C. 2007. Corpora for SLA: using Sketch Engine
to learn Chinese and English. Proceedings 2007 Conference and Workshop on TEFL
and Applied Linguistics. pp 430-436. Taoyuan.
Smith, S., Kilgarriff, A., Gong W, Sommers, S., Wu G. Forthcoming. Automatic Cloze
Generation for English Proficiency Testing. Paper to be presented at 2009 LTTC
International Conference on English Language Teaching and Testing. Taipei.
Vanessa Metcalf, V. & Meurers, D. 2006. Generating Web-based English Preposition
Exercises from Real-World Texts. Proceedings of EUROCALL 2006. Granada, Spain.