Improving Computational Tools for Ubiquitous Online Collocation Learning

1
Improving Computational Tools for Ubiquitous
Online Collocation Learning
Second International Roundtable on Digital
Language Learning Integrating Design and
Practice

• Anne Li-Er Liu

2
What is collocation?

• Van Roey (1990)the linguistic phenomenon
  whereby a given vocabulary item prefer the
  company of another item rather than its synonyms

• Sinclair (1991)the occurrence of two or more
  words within a short space of each other in a
  text

• Lewis (2000132)the way in which words co-occur
  in natural text in statically significant way

3
MiscollocationsVerb Noun pairs

• MiscollocationsAn incorrect word combination
  that breaks the convention of a collocation.

I know as long as I can take effort in what I
want, I will have a happy life.
appear his fear
make a conclusion
4
UWiLL--ubiquitous web-based interactive language
learning

• Wible et al. (2005) have designed a browser-based
  agent to present a revolutionary collocation
  learning approach.

UWiLL Collocator
5
Learning while surfing?Surfing while learning?

• surfing on the Internet

content
language
meaning
form
6
UWiLLCollocator

• Collocation Retrieval Approach

A modified MI approachMIgt1
1. Statistical-based
2. word span 5 words
(5 words window)
7
take
medicine

8
pen
take

take medicine, MI 1.5849
9
Collocation Retrieval Approach

• The modified MI

5 words window
?
?
?
? semantic-independent approach
Its going to take time and the medicine is not
working.
?
10
UWiLL Collocator
11
(No Transcript)
12
Computational Approach vs. Linguistic Approach in
retrieving VN collocations

• Computational view

?
precision
13
Computational Approachvs. Linguistic Approach in
retrieving VN collocations

• Linguistic view

Within the 5 words window,
1. Verbs/ nouns are not allowed.
2. Certain punctuation is not allowed.
High precision
?
Recall
3. Certain words are not allowed.
4.
14
A refined approach in retrieving collocations on
the web

• Adjust MI value from 1 to 3.

• Constraints inside the 5 words window

• ? Verbs and Nouns

It is going to take time and the medicine is not
working.
? Punctuation
Mary didnt run the race was canceled.
? Other words conjunctions, prepositions.
She couldnt tell and the secret was kept
15
(No Transcript)
16

• PUT screen dumps here with old version where
  punctuation is not removed.

17
seek comfort is removed.
18

• PUT screen dumps here with old version where VN
  are not removed.

19
plan to alter its strategy
20
(No Transcript)
21

• PUT screen dumps here with old version where some
  words are not removed.

22
(No Transcript)
23
After adding one retrieving criterion
24
Discussion

• Close collocations tend to have higher MI value
  when the focal word or the collocate is a high
  frequency word, or both are high-frequency words,
  their MI value tends to be significantly lower.

How do we deal with high frequency words in a
collocation?
take medicine MI 1.5849
make coffee MI1.7648

• MI 3 or 1?

put head MI2.5564

• Passive voice

• Removing verbs/nouns seems not help much in terms
  of increasing precision.

25
(No Transcript)
26
Discussion

• Close collocations tend to have higher MI value
  when the focal word or the collocate is a high
  frequency word, or both are high-frequency words,
  their MI value tends to be significantly lower.

How do we deal with high frequency words in a
collocation?
take medicine MI 1.5849
make coffee MI1.7648

• MI 3 or 1?

put head MI2.5564

• Passive voice

• Removing verbs/nouns seems not help much in terms
  of increasing precision.

27
Collocation Project
Find 5 collocations
In-class discussion
Read a news story
Turn them in
Report finding in IWiLL
Hands on IWiLL Collocator
Use UWiLL Collocator outside the class
Comparison
28
School Bullying in Japan
In Japan two more children have killed themselves
after being bullied at school. 5 children have
taken their own lives since August. A headteacher
has also been found hanged in a forest after he
was criticised for his response to a bullying
incident in his school. Both of the children
who killed themselves this weekend had been
bullied. A twelve year old girl jumped to her
death from a high building after her classmates
teased her for being short. A fourteen year old
boy hanged himself after his classmates tried to
extort money from him. It's not new of course.
Japan's schools have long had a problem with
bullying. After the suicides of five young
children in less than four months, the society
has paid attention to the problem, but no-one can
agree what to do about it. More attention, of
course, can create its own difficulties. The head
teacher killed himself after giving a news
conference in which he was criticised for his
handling of an extortion incident involving a
pupil at his school. Committing suicide in Japan
is not just seen as an easy way out.
Traditionally, it was regarded as a responsible
way to deal with a problem when faced with no
other choices. But last week, several
schoolchildren wrote to the Education minister
threatening to kill themselves because no-one
would help them. He responded by urging students
not to write to him, but instead to try find
people they could talk to about their
difficulties. It's awkward. Some fear
publicising such cases increases the likelihood
others will try to copy them. Others though, say
the failures in the system need to be exposed.
29
Students collocations, 10 groups

• increase likelihood

? 1

• kill himself/themselves

? 3

• deal (with) problem

? 6

• create difficulty

? 6

• hang himself

? 2

• give conference

? 5

• pay attention

? 7

• commit suicide

? 2
30
(No Transcript)
31
(No Transcript)
32
Special Thanks to

• Nai-lung Tsao Computer and Network Lab, Tamkang
  University
• Chong-Fu Hong Computer and Network Lab, Tamkang
  University