WordGap Automatic Generation of Gap-Filling
Vocabulary Exercises for Mobile Learning
Susanne Knoop
Sabrina Wilske
University of Bremen, Germany
Digital Media in Education (Prof. Heidi Schelhowe)
May 22, 2013
Susanne Knoop (University of Bremen)
WordGap
May 22, 2013
1 / 22
Overview
1
Motivation
2
Previous Work
3
WordGap
4
NLP Tools
5
Outlook
Susanne Knoop (University of Bremen)
WordGap
May 22, 2013
2 / 22
Obstacles To Language Learning
Susanne Knoop (University of Bremen)
WordGap
May 22, 2013
3 / 22
The Solution
1
Free CALL software
2
Mobile apps (learn everywhere, everytime)
3
Exercises from texts that are of special interest to the user
Susanne Knoop (University of Bremen)
WordGap
May 22, 2013
4 / 22
Vocabulary Learning With Multiple-Choice Cloze Exercises
ˆ
ˆ
Cloze exercises consist of:
1
A sentence
2
A target word that has been replaced by a gap
3
(normally three) distractors to distract from the target response
They are widely used in second language training and testing
Susanne Knoop (University of Bremen)
WordGap
May 22, 2013
5 / 22
The Challenge
ˆ
It normally requires expert knowledge to create cloze exercises:
1
Distractors of multiple-choice cloze exercises should not be totally
unlikely but also not too similar to the target word
2
Distractors should have the same grammatical form as the target word
Susanne Knoop (University of Bremen)
WordGap
May 22, 2013
6 / 22
Previous Work
ˆ
Coniam (1997), Brown et al. (2005): Distractors of the same part of
speech and similar frequency as the target word
ˆ
Sumita et al. (2005): Distractors are false synonyms of the target
word (veried by a Google search)
ˆ
Evaluation: Performance of students in the automatically generated
tests correlates strongly with standard vocabulary tests.
Susanne Knoop (University of Bremen)
WordGap
May 22, 2013
7 / 22
WordGap
ˆ
server component and mobile app for Android smartphones
ˆ
informal vocabulary learning, adapted to the learner's interest
ˆ
multiple-choize cloze exercises from a text le or webpage
Susanne Knoop (University of Bremen)
WordGap
May 22, 2013
8 / 22
Share a webpage from the browser
Susanne Knoop (University of Bremen)
WordGap
May 22, 2013
9 / 22
Send the webpage to WordGap
Susanne Knoop (University of Bremen)
WordGap
May 22, 2013
10 / 22
Choose a part of speech
Susanne Knoop (University of Bremen)
WordGap
May 22, 2013
11 / 22
Display the exercise sentence by sentence. . .
Susanne Knoop (University of Bremen)
WordGap
May 22, 2013
12 / 22
. . . receive immediate feedback. . .
Susanne Knoop (University of Bremen)
WordGap
May 22, 2013
13 / 22
. . . receive immediate feedback. . .
Susanne Knoop (University of Bremen)
WordGap
May 22, 2013
14 / 22
. . . get the result.
Susanne Knoop (University of Bremen)
WordGap
May 22, 2013
15 / 22
The server-client model of WordGap
Susanne Knoop (University of Bremen)
WordGap
May 22, 2013
16 / 22
The NLP pipeline of the WordGap server
Susanne Knoop (University of Bremen)
WordGap
May 22, 2013
17 / 22
[plain]
Susanne Knoop (University of Bremen)
WordGap
May 22, 2013
18 / 22
WordNet
ˆ
Hierarchic semantic network of the English language (Miller and
Fellbaum, 2007)
ˆ
WordGap: nd false synonyms or antonyms of a target word as
distractors
ˆ
False synonyms: synonyms of synonyms or words with the same
hypernym
ˆ
Distractors from the text's context often more suitable than those
from WordNet
Susanne Knoop (University of Bremen)
WordGap
May 22, 2013
19 / 22
Natural Language Tool Kit
ˆ
extensive open-source Python library for NLP
ˆ
WordGap: sentence and word tokenizing and part of speech tagging
ˆ
Naive Bayes Tagger with a cut-o probability of 0.95
Susanne Knoop (University of Bremen)
WordGap
May 22, 2013
20 / 22
NodeBox Linguistics
ˆ
collection of open-source Python libraries for NLP
ˆ
WordGap: generate dierent grammatical forms to adapt distractors
to the target word
Susanne Knoop (University of Bremen)
WordGap
May 22, 2013
21 / 22
Outlook
ˆ
Include grammatical exercises, e.g. dierent verb tenses
ˆ
Extend to other target languages
Susanne Knoop (University of Bremen)
WordGap
May 22, 2013
22 / 22