The Technology

 

A standard speech recognition system is adapted to allow robust recognition of accentuated and erroneous speech data. Systematic detection and evaluationof learner errors form the basis for the diagnostic and feedback modules.

 

The ISLE Prototype System

An initial prototype system will be produced with the following architecture.

Top Level

The top-level language tutoring system is responsible for working directly with the student, by choosing and presenting exercises, assessing the accuracy and quality of the student’s spoken responses (by calling the lower-level functions), and providing helpful feedback. Because of the modular design of the lower-level systems, a language teacher will be able to add new exercises and exercise types without having to understand or have access to the low-level functions themselves.

Mid Level

The mid-level ISLE „Monitor“ provides a simple but flexible interface to the powerful low-level functions that lie below it. The monitor is responsible for converting the requests from the top-level into the specific calls needed for the low-level functions, thus insulating the top-level designer from the complex details.

Low Level

Four low-level modules form the core of the ISLE system. The first module recognizes the utterance spoken by the student—despite possible mispronunciations—with a very high degree of reliability. The second module analyzes the signal, looking for regions that have been poorly pronounced. The third module decides what kinds of phonological errors have been made, for example, substitutions of one phoneme for another. The final module analyzes word-stress, so that the system can provide feedback for these highly prominent errors.