... is:¹

We omit enumeration of the function e but it could be given in a similar manner, for example ((1,2), agent) is an element of e.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... substrings²

We consider two substrings to be the same string if they contain the same characters in the same order, irrespective of their positions within the larger string in which they occur.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...),³

We omit the subscript on M when the set STF is obvious from context.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... 1996⁴

In fact, all of these assumptions except for single-use were made by siskind:cog96; see Section 7 for details.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... meanings.⁵

These words may, however, serve as cues to a parser on how to assemble sentence meanings from word meanings.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... fractures.⁶

Thanks to net-citizen Dan Hirshberg for help with this analysis.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...⁷

The code is available upon request from the first author.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... lexicon.⁸

Though, of course, interpretation functions are not the only way to guarantee a covering lexicon - see siskind:ircs93 for an alternative.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... learner,⁹

The predicate is omitted because CHILL initializes the parse stack with the answer predicate, and thus no word has to be mapped to it.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... here,¹⁰

The CPU times of the two system are not directly comparable since one is written in Prolog and the other in Lisp. However, the learning time of the two systems is approximately the same if Siskind's system is run in incremental mode, just a few seconds with 225 training examples.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... constructed.¹¹

Thanks to Jeff Siskind for the initial corpus generation software, which we enhanced for these tests.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...¹²

In these tests, we allowed WOLFIE to learn phrases of up to length two.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.