A CommonLisp Interface to WordNet

About WordNet

Professor George Miller of the Cognitive Science Laboratory of Princeton University directed the development a lexicographic database called WordNet.

Princeton maintains a server by which the WordNet database can be browsed via the World Wide Web.

The WordNet database is implemented as a set of text files. Mark Nahabedian (naha@ai.mit.edu) has developed an interface to this database written in CommonLisp. This software provides an interface by which CommonLisp programs can access lexicgraphic data from WordNet.

CommonLisp Interface

The interface is written in several layers:

There is also a simple browser implemented in CLIM for navigating the WordNet database.

This software represents parts of speech as lisp keyword symbols: :noun, :verb, :adjective and :adverb.

The current version of this software only knows how to find WordNet index and data files as they are named in the UNIX implementation of WordNet. Set the value of the parameter wn::+wordnet-database-directory+ in the file "wordnet-database-files.lisp" to the pathname of the directory where these files can be found.

The current version has only been tested with Symbolics Genera and Macintosh CommonLisp (thanks to Andrew Blumberg, blumberg@ai.mit.edu). The software might require slight modification to run on other Lisp Implementations.

All the files can be found here. A single file in UNIX tar format is also available.

The Base Layer

The base layer defines the packages and export lists for this software. It is implemented by these files:

Record Extraction

 The record extraction layer is the bottom-most one. It implements functions which extract records from the database files as text strings.
(index-entry-for-word file-description word)
Looks up word in the specified index file and returns the string corresponding to that record of the index file. The file-description argument can either be a part of speech keyword, a pathname naming an index file, or a stream which has been opened to that file.
(read-data-file-entry file-description offset)
Reads a WordNet "symset" record from the specified offset in the specified file. A string is returned. Offset was either read from an index record, or from a pointer description in another synset record. The file-description argument should identify a WordNet data file. It should either be a part of speech keyword, a pathname, or a stream.

This layer is implemented by the file wordnet-database-files.lisp.

This layer depends on the files in the base layer.

Record Parsing

The functions in this layer take strings as returned by the functions of the record extraction layer. They parse those strings into components, returning them as multiple values.
(parse-index-file-entry entry)
Parse the entry as returned by index-entry-for-word. See the definition for a list of the values returned.
(parse-data-file-entry entry)
Parse the entry as returned by read-data-file-entry. See the definition for a list of the values returned.

This layer is implemented by the file parse-wordnet-data.lisp.

This layer depends on the files in the base layer.

Data Representation

The data representation was chosen to parallel WordNet's own representation. It models index entries, synonym sets and pointers. Depending on ones application, there might well be more useful ways to represent the WordNet lexicon. Practice might lead us to modify this representation or develop a new one.

Class wn:wordnet-index-entry
Objects of this class are used to represent entries read from the index files. They are created and returned by the function wn:cached-index-lookup.
(wn:cached-index-lookup word part-of-speech)
Looks up word in the index file corresponding to part-of-speech and returns an index entry object for it.
(wn:index-entry-synsets index-entry)
Returns a list of the synonym sets, as wn:wordnet-synset-entry objects, which index-entry refers to.
Class wn:wordnet-synset-entry
Objects of this class represent synonym sets. There is a subclass for each part of speech:
(wn:synset-words synset)
Returns a list of "words" that are in the synonym set synset. Each word is represented by a list, the first element of which is the word as a string. The second element is the sense number assigned by the lexicographer.
(wn:wordnet-pointers synset)
Returns a list of the wordnet pointers from the specified synset.
Class wn:wordnet-pointer
These are how wordnet pointers are represented.
(wn:wordnet-pointer-type pointer)
Returns the wordnet pointer type for pointer, e.g. :antonym, :hypernym, :entailment, etc.
(wordnet-pointer-from-synset pointer)
Returns the synonym set which pointer points from.
(wordnet-pointer-to-synset pointer)
Returns the synonym set which pointer points to.
(wordnet-pointer-from-word pointer) (wordnet-pointer-to-word pointer)
If pointer refers to a specific word in the synonym set, that word (as a list of string and sense number) are returned, otherwise the synonym set is returned.

This layer is implemented by the file representation.lisp.

This layer depends on the files in the base layer, the record extraction layer and the record parsing layer.

Pointer Reasoning

This layer provides some functions for operating on the graph formed by WordNet synonym sets and the pointer relationships among them. Here follows a description of the operations currently provided. This set is expected to grow with time.

(wn:relation-transitive-closure synset relation-type)
relation-type must be a WordNet pointer type representing a transitive relation. This function returns a set which is the transitive closure of that relation starting with synset. The closure set is returned as a list. Each element of the list is a cons whose car is a synset object and whose cdr is an integer rpresenting the distance along the relation-type between this synset and synset.
(wn:commonality relation-type &rest synsets)
Finds the common "ancestors" of the synset objects in synsets along the relation-type graph. It returns a list, the first element of which is the closest common ancestor. The rest of the list has one element for each of synsets. Each element is a cons whose car is one of the synsets and whose cdr is the distance from this synset to the common ancestor.

This Layer is implemented by the file relationship-algorithms.lisp.

This layer depends on the data representation layer.

Browser

The browser provides a simple user interface for examining the wordnet database. It defines CLIM presentation types and commands for displaying the objects defined in the data representation layer.

It depends on the data representation layer, and on the layers on which that layer depends.

The browser also depends on a domonstration lisp interactor implemented in CLIM, which in the Symbolics Genera CLIM distribution can be found in the directory "sys>clim>rel-2>demo>listener.lisp".

The command :Lookup takes a string as argument. It looks up that string in the indices and prints out a list of index entries that were found.

You can click on one of these index entries to get a list of the synonym sets that it refers to.

Clicking on a synonym set will list the pointer references that it has to other synsets. The presentation of the pointer includes a presentation of the synset that it points to. You can click in it in turn to see its pointers.

Examples

Some examples have been written which illustrate the use of this software. Included are functions which list synonyms and antonyms for a specified word, and a function which lists the names and nicknames of the U.S. States. There is also a function which tries to identify the synset for a word having a sense most similar to a specified word by comparing distances along hypernym pointers among the synsets for the word being looked up and the sense indicating word.