CHAT Research Notes

3/10/2000

From proposal:

Figure 1. Architecture for an inquiry-based tutor

Revised:

observer agents:

• link parser, produces list of probable failures
• grammatical difference analyzer
• lexical difference analyzer
• lexical semantic consistency checker, flags anomalies in lexical markings inferred from semantics (wordnet) (eg consider synonyms with different feature settings)

reporter agents:

• critic (report problematic sentences)
• teacher (mini-lessons, suggest new exercises)
• spy (see how others differ)
• judge (assess strengths and weaknesses, praise success)

Note that while most reporters get information primarily from one observer, information from other observers can provide contextual information etc.

Agents run automatically or user invoked? Observers automatically, reporters invoked (and maybe sometimes automatically too). A very rough sketch of a possible menu: