9561577 Bernstein The assessment of spoken language proficiency is essential for prospective professions (such as nurses or flight attendants) if they are not native speakers of English but need to work in situations where rapid, accurate communication is a vital concern. Assessment of spoken language proficiency can also figure in the qualification for many other professions, for example for teachers. This Small Business Innovation Research Phase I project will investigate the use of speech recognition in language testing. The guiding objective is to demonstrate techniques that suggest that a feasible system for Automatic Spoken Language Assessment by Telephone (ASLAT) can produce proficiency scores that are reliable and valid with reference to good current tests. Phase I research will implement a telephone-based system for pilot testing various interactions with populations of native and non-native speakers. The immediate objective is to identify several interactive tasks that exercise a range of receptive and productive language skills which reliably predict a test-taker's level of functional oral proficiency in English. These automated interactions should, first of all, be quite simple for native speakers of North American English. That is, regardless of dialect or other demographic characteristics of a native speaker, an ASLAT system should return a high score. At the same time, the ASLAT should distribute non-native test-takers over a wide range of scores that reflect the range of their functional command of spoken English. There is good reason to believe that advanced speech recognition systems adapted for language assessment and operating over the telephone will enable the achievement of the project goal of offering valid and reliable spoken language assessments at low cost the user. Current tests of spoken language proficiency are usually administered, scored, and reported by human operators. The labor cost to administer a typical test ranges from $10 to $60. Computer-b ased tests of oral proficiency can probably be administered, scored, and reported for between $1 and $2 per test, leaving another $3 or $4 per test in margin when tests are offered at $5 per test in large volumes.

Project Start
Project End
Budget Start
1996-02-01
Budget End
1996-07-31
Support Year
Fiscal Year
1995
Total Cost
$72,544
Indirect Cost
Name
Entropic Research Laboratory Incorporated
Department
Type
DUNS #
City
Menlo Park
State
CA
Country
United States
Zip Code
94025