Investigating Chinese Learner English

1
Investigating Chinese Learner English

• Centre for Linguistics and Applied Linguistics,
• Guangdong University of Foreign Studies
• Gui Shichun

2
Background

• The corpus consists of one million words of
  written compositions by 5 types of learners
  senior middle-school, tertiary college English
  (band 4), tertiary college English (band 6),
  tertiary majors in English (1st and 2nd years),
  tertiary majors in English (3rd and 4th years).
  The corpus is annotated with grammatical tags
  (automatically) and error tags (manually).
• It is avaiblable at http//www.clal.org.cn/baseinf
  o/ achievement/ Achievement1. htm,

3
(No Transcript)
4
Areas of Investigation

• Leech (1998) raises two specific questions in
  connection with the study of learner language
• What are the particular areas of overuse,
  underuse and error which native speakers of
  language A are prone to in learning target
  language T, as contrasted with native speakers of
  languages B, C, D . . . ?
• What, in general, is the proportion of non-native
  target language behaviour (overuse, underuse,
  error) peculiar to native speakers of language A,
  as opposed to such behaviour which is shared by
  all learners of the language, whatever their
  mother tongue?

5

• Contrastive study must be very careful, because
  the corpora under investigation are based on
  different types of language performance. One of
  the key issues is to identify the context-free
  variables, e.g. functional words and some of the
  most frequently used notional words. We believe
  that an annotated learner corpus is useful in the
  following ways

6

• Identifying the words and structures that are
  typically underused or overused in the learner
  corpus
• Identifying the kinds of error learners at
  different levels are likely to commit
• Predicting the language proficiency of the
  learners
• Providing diagnostic information to the both the
  teachers and the learners.

7
Comparison of grammatical tags

• In our POS tagging program, we used the same
  133133 matrix of tag-transition frequencies, and
  had CLEC grammatically tagged automatically. Then
  we tried to compare the grammatical tags of CLEC,
  Brown, and LOB.

8

• The native corpora (BROWN and CLEC) are fairly
  consistent in terms of their grammatical tagging.
• Chinese learners used more pronouns, but fewer
  determiners, prepositions, and numerals. Use of
  more pronouns and fewer numerals reflects the
  differences of subject matter between the learner
  corpus and the native speaker corpus, because
  what the learners have written are related to
  their personal and school life and activities.
  But use of fewer determiners and prepositions may
  have something to do with the learner problems in
  their writing.
• Another step forward is to study in greater
  details each type of the tagging scheme. Lets
  look at use of determiners as an example.

9
(No Transcript)
10
(No Transcript)
11
Some observations of Chinese learners use of
determiners

• Chinese learners used fewer determiners, but the
  total frequencies of ST6 learners were closer to
  those of the native speakers.
• Chinese learners used fewer articles the, no
  a, any were the most underused determiners.
• Some tendency can be observed the more
  proficient the learners, the closer is their use
  of some determiners to that of the native
  speakers. For example, quite, rather, half, all,
  both, these, those, many, much, next, former, and
  other. The last five are post-determiners, which
  were used much more often by native speakers.
  They can be considered as discourse markers. We
  hypothesize that they can be used for text
  identification as automatic scoring of learners
  compositions.

12
Under-use of the learner lexicon

• The most frequently used lexical items are
  more or less context-free, and it is a suitable
  place to start with our analysis. They include
• Use of most functional words like determiners and
  prepositions
• Some of the modal or auxiliary verbs
• Some of the polysemous words like go, make, take,
  great, risk, etc
• Some pronouns, especially personal pronouns.

13
Overuse of Modal Verbs (can, may,much,should)
14

• As observed by Biber et al , can (ability or
  permission), must (logical necessity) are used
  much more common in conversation, the overuse of
  can and must can be considered as an indication
  of Chinese learners writing style. They make no
  distinction between the stylistic differences of
  spoken and the written forms. The materials of
  CET learners were collected mostly from CET test
  papers, yet they displayed the greatest number of
  uses of can, must, and should. Chinese learners
  tend to write down what they speak, though they
  may not be well versed in speaking, as is
  indicated by the underuse of could, have to, had
  better and have got to.

15
Comparing keyness of CLEC and FLOB

• By using the keyness programme of Wordsmith, we
  are able to identify the underuse of the learner
  corpus in terms of keyness,which is the classic
  chi-square test of significance with Yates
  correction for a 2 X 2 table. For better
  estimation of keyness, Ted Dunning's Log
  Likelihood test is used when contrasting long
  texts or a whole genre against the reference
  corpus. The higher the chi-square value, the
  greater is the difference between the frequencies
  of two corpora under observation.

16
Fewer third person pronouns

• This is the result of Chinese transfer,
  because in modern colloquial Chinese, the third
  person pronouns do not make any gender
  difference. There is no underuse of first and
  second person pronouns. St3-4 learners show wider
  discrepancies, because their compositions are
  mainly thematic writing.

17
Fewer passive voice constructions

• This is shown by the underuse of been and by,
  and partially by was and were. The st3-4
  group and the st5-6 group seem to follow the same
  tendency

18
Fewer relative clauses

• Chinese learners tend to use fewer relative
  clauses, as is shown by the underuse of wh-words.
  The discrepancies of st5-6 are smaller, showing
  that they are closer to the native speakers.

19
Contrastive analysis of risk and its synonyms
across a few corpora
Using more danger than risk
20

• While the frequencies of risk, danger, threat,
  and hazard are fairly consistent in the native
  speaker corpora, the performance of Chinese
  learners is quite different.
• Danger is a more generic term. The following
  errors are produced as a result of the generic
  use of danger
• Fake furniture brings danger to people.
  (It is risky buying fake furniture.)
• Water is facing the danger of shortage.
  (We are facing the threat of water shortage.)
• Their knowledge of risk is quite limited. They
  know how to use take the risk(8), at the
  risk(3)and to risk(6) whereas native speakers
  say avoid/carry/eliminate/ignore/crease/involve/
  give/reduce/run/ worth/lack of / the risk
  conventional/maximum/no/some/suicide/own/
  unnecessary/hazard/ with/ without/ risk
• Chinese learners do not know how to use high risk
  ,which is used quite often in the native speaker
  corpora?

21
Analyzing learner errors The Cognitive Model

• We use error as a cover term for all ways of
  being wrong as an FL learner. Errors are results
  of uncertainty in language performance, and
  there are various kinds of uncertainty that can
  be traced back to cognition
• False analogy books, news
  knowledges, informations
• Incomplete application of rules
  development advantagement
• Redundancy ??????????it was a
  three-story-tall building
• Overgeneralization entered the
  classroomreturned the classroom

22

• Verbal behaviour (errors as well as linguistic
  structures) can be considered as an emergence
  process, as a result of competition of cues.
• To set up our cognitive framework of error
  analysis we make use of only those errors whose
  frequencies are well above 1 of the total. There
  are altogether 21 error types.
• Errors can be divided into several levels that
  are equivalent to the processes of lexicalization
  ? syntaticalization ?relexicalization
  (Skehan,1998).

23

• Lexical perceptual level, also known as
  substance errors (James, 1998), and defined by
  MacWhinney as the level that involves the
  acquisition of basic lexical structures in small
  areas of cortex called local maps. They are
  related to perceptual representations, especially
  to memory, such as memory failure or memory
  distortion. Typically these errors can be
  identified at single-word level, as
• spelling or number errors (great graet
  information informations) ,
• or by looking at its close neighbors as
• absence of articles or prepositions. (the moon
  is?brightest I dressed myself in?hurry I sat
  back?my chair).

24

• Lexico-grammatical (or lexical grammatical)
  level. Misconception of target language system.
  When looking at the errors of our learners, it is
  very difficult to isolate grammar and lexis into
  separate categories, because grammar does not
  exist on its own. James defines it as text-level
  errors whereas MacWhinney chooses to call it
  the level that involves the interaction between
  lexical structures in terms of lexical groups.
  Typically these errors can be identified at the
  inter-word level, by looking at the word and its
  neighbors.
• using the wrong parts of speech (POS errors It
  is not difficulty that we can find)
• wrong word (substitution errors ???????If
  you match difficult problemPeople take
  (pay)more attention to it)
• wrong collocates (They must listen to the lesson
  more carefully.)
• verb agreement (People argues that euthanasia or
  mercy killing is humane.)
• Reference (My aunt came to my home with his
  son.).

25

• Syntactic level. Errors can be identified at a
  broader context, at the sentential level. James
  chooses to call it discourse-level errors, but
  we propose to reserve the word discourse for
  another upper level. L2 learners may often
  produce grammatical sentences that sound foreign.
  (Pawley and Syder, 1983). MacWhinney defines it
  as the level that involves the processing of
  syntactic information across longer neural
  distances in functional neural circuits.
  Syntactic errors vary from

26

• capitalization (he learned English and Russian
  and Wrote the Civil War in France. )
• punctuation (When playing football or basketball.
  You might be using 400 calories an hour.), to
• run-on sentences (If I am not famous, it doesnt
  matter, I dont mind this.),
• fragmentary sentences (As they do more exercises
  and often think deeply.) and
• structural deficiency (During I spent my holidays
  in Beijing about ten years ago,).

27

• Figure 9 The Cognitive Model

28

• Confirmatory factor analysis was conducted by
  using Lisrel 8.50, which shows clearly that there
  are 3 factors, and they are grouped under 3
  categories as what have been defined. Path
  analysis shows that all the parameters (values of
  ?s) of the hypothetical paths are significant
  except run-on sentences.

29
Lamda0.28, insignif-icant
30
Correspondence Analysis (learners types by error
types)
31
Some General Remarks

• On the whole, identifying errors at 3 levels
  seems to be working well in our cognitive model.
  So far weve not covered errors at the discourse
  level, because,
• It is difficult to set down the standards for
  native-like selection as defined by Pawley and
  Syder (1983)
• It is even more difficult for Chinese markers of
  errors to observe the standards.

32

• The grouping of errors is not as clear-cut as
  what weve thought. Very often the same type of
  error can be put into different categories or the
  same type of errors can occur at 3 different
  levels depending on the situations. We can only
  say this is done according to the main tendency.
• At every level, language transfer seems to play
  an important role. This is because the adult
  learners have set up their L1 (more complete)
  linguistic system and are in the process of
  setting up another linguistic system (rather
  incomplete). As mature learners, when they want
  to express their complex thinking, they often
  fall back on using the linguistic system that is
  more familiar to them.

33

• Occurrences of errors depend very much on the
  writing task and the learners certainty of
  fulfilling the task. They may not be an
  indication of the language proficiency of the
  learners. CET learners tend to commit more
  lexico-grammatical errors because their data were
  collected mainly from CET compositions.
• This points to the necessity of inclusion of more
  learner data, so that we can have a more balanced
  collection of error types for further
  investigation.

34
Thanks!
Your Comments are Welcome!