> I'm playing around with word frequency lists to generate test items of
> similar word frequency.
>
> Thus say you want to test a word like "country" which is frequency 127
> - i.e. the 127th most frequent word in English ("the" is frequency 1)
> in the Bank of English's tagged wordlist, you might end up with the following
> (the number after each word is the word's respective frequency)
>
> government 100
> state 124
> thing 125
> country 127
> man 158
>
> So with a word like "conviction" (freq. 4928), you'd get
>
> plea 4905
> copper 4910
> presentation 4911
> sigh 4912
> conviction 4918
>
> First reactions are that *not* choosing semantically-related words to sample
> language proficiency must be wrong, altho there is research which suggests a
> relationship between control of the (more and less) frequent words and
> proficiency (Harlech-Jones, 1983; Meara and Jones, 1988)
>
> Having trialed a couple of tests derived in such a way with students,
> I didnt get total GIGO as I'd half been expecting.
>
> Is anyone else working along similar lines, or wd like to comment?
>
> David Coniam
> Chinese University of Hong Kong
> coniam@cuhk.hk
This is is a topic I'm also very interested in and is close to the core of my
Ph.D. work. Whilst it would be a gross over-simplification to claim that
frequency is the only criterion that would dictate the acquisition of a
lexical item, there seems to be a common sense appeal that it is likely to have
a major impact. Studies in L1 learning seem to bear this out, and I see no
reason why it shouldn't be the case for L2 learners. There are obviously
plenty of caveats to what I've said.
I'd also be interested in hearing from anyone who is using corpora for L1 or
L2 studies of lexical acquisition and I'm sure that those who are engaged in
NLP research of the lexicon will be able to offer some insights from their
work.
John Shillaw
Foreign Language Center/Institute of Modern Language & Culture
University of Tsukuba, Japan
e-mail: jds@sakura.cc.tsukuba.ac.jp