... categories¹

Natural categories are those for which people can readily associate typical members. For instance, CHAIR would be a natural category instead of FURNITURE. Natural categories are also referred to as basic-level categories, because they tend to occur at the level in a taxonomy where most of information resides, in terms of attributes [103,102].

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

... languages.²

The EuroWordNet project [119] is seeking to tie together separate ``wordnets'' that are being developed for several different languages. It is addressing the problem of having separate ontologies for each language by specifying high-level correspondences.

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

... test:³

The likelihood is generally defined as the conditional probability of the data (or evidence) given the hypothesis, which would be $P(v, n\ \vert\ verb\_attach\ p)$ for the verb case. Via Bayes's Theorem, this can be rewritten $P(verb\_attach\ p\ \vert\ v, n)P(v, n)$ and, since the prior probabilities cancel in the likelihood ratio, the form given above results. Logarithms are often used with likelihoods ratios (hence the log-likelihood test) to make combinations additive rather than multiplicative[85].

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

... text⁴

This co-occurrence matrix reduction algorithm is similar to that used in latent semantic analysis    [33,42], which is not discussed in this review.

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

... alternation.⁵

The dative alternation licenses ``X give Y to Z'' $\iff$ ``X give Z Y'.

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

... expressions⁶

This uses the following conventions: X matches category X; . matches any category; X* matches 0 or more X's; X? matches 0 or 1 X's; and, $\langle Z\rangle$ applies rule Z

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

... parser⁷

The Microsoft English Grammar (MEG) parser is based on the PLNLP English Grammar (PEG) parser developed at IBM (PLNLP stands for Programming Language for NLP).

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

... function⁸

Zipf's law states that a term's frequency is inversely proportional to its rank (e.g., frequency of third most common term is one-third that of the first). The curve plotting this relationship can be viewed as the top half of a hyperbola (rotated 45 degrees).

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