Screen shots of IToldU

3
Bilingual Lexical Data Contributed by Language Teachers Via a Web Service: Quality vs. Quantity V. Bellynck, Ch. Boitet, J. Kenwright INGP & UJF, Grenoble, France A knowledge engineering context Three questions in such a context General goal: collect detailed lexical information about many languages The knowledge to build is HIDDEN, IMPLICIT hence comparable to search for exoplanets but it HAS TO BE ELICITED FROM HUMANS In order to build a sizable amount of reliable knowledge, - WHO will contribute (expertise, etc.) ? - WHAT kind of information will be built ? - HOW to elicit contributions ? voluntary & opportunistic / organized constrained (hence organized)

description

Bilingual Lexical Data Contributed by Language Teachers Via a Web Service: Quality vs. Quantity V. Bellynck, Ch. Boitet, J. Kenwright INGP & UJF, Grenoble, France. Screen shots of IToldU. Dictionarial evaluation (first year). - PowerPoint PPT Presentation

Transcript of Screen shots of IToldU

Page 1: Screen shots of IToldU

Bilingual Lexical Data Contributed by Language Teachers Via a Web Service:

Quality vs. Quantity V. Bellynck, Ch. Boitet, J. Kenwright

INGP & UJF, Grenoble, France

A knowledge engineering context Three questions in such a context

General goal: collect detailed lexical information about many languages

The knowledge to build is

HIDDEN, IMPLICIT

hence comparable to search for exoplanets

but it HAS TO BE ELICITED FROM HUMANS

In order to build a sizable amount of reliable knowledge,

- WHO will contribute (expertise, etc.) ?

- WHAT kind of information will be built ?

- HOW to elicit contributions ?

voluntary & opportunistic / organized

constrained (hence organized)

The IToldU project offers an effective combination

Page 2: Screen shots of IToldU

Screen shots of IToldU

Invented example

Error: the teacher willoverstrike it

Page 3: Screen shots of IToldU

Dictionarial evaluation (first year)

Quantitative aspect Qualitative aspect (base on the revision of first year contribution)

First semester : 12,000 English-French entries were entered by the students, along with about 8,000 usage contexts

Second semester : +5,000 English-French entries with less than 200 entries lacked contexts

After the first year of use to increase the free usage context contributions, the number of entries to contribute has been limited to 2*30 per student for technical field and for common vocabulary

Second year : +4000 entries

Third year, first semester ; +2000 entries

Apart from errors arising from problems in inputting diacritics on the Web, the French translations of English terms are almost all correct. By contrast, 15% to 20% of usage contexts are not examples of use.

Translations

95% of the translations seem correct to us,

about 30% concern a purely technical lexical field, while 70% concern “paratechnical” fields,

From usage contexts to examples of us

Teachers evolves from asking usage context to examples of us, because of -several unexpected contributions such as telling the « domain » of a citation, or a kind of définition (without using the word to illustrate), -emerging king of « invented » context