CS365

Homework 0

Psychotherapy sessions with Eliza

The ELIZA chatterbot snippets:

A1 - (where is works “somewhat” reasonably)

The ELIZA code has large data source where it stores and classifies large number of english language words. ( Following is snippet of ELIZA code with a verb function and other is classified words). It also has a history buffer.

The doctor tries to identify keywords, subject, verb and nouns from the sentence. The KEY is usually stored in history for asking questions in future :

The doctor also calls a transposition function to change your questions into relevant questions:

   //         “of” transposed to “by”

But this doesn’t always work well.  For instance consider the following snippet :

A2 - (poor responses)

Apart from this there are various standard responses that have been included in the program drawing from a Rogerian Psychotherapist Expert System (source: Wikipedia).

Scope for Improvement:

• The rogerian responses are many time random. A better context detemination scheme to pick responses.
• The transposition fails many times. More involved constuction rules may be integrated for English language.
• The database of words and the classification is severely restricted for a language as vast as English.