Greg Detre
Tuesday, October 01, 2002
1. Summarize the main ideas of each paper (one
page maximum for each summary).
2. Download WordNet
(http://www.cogsci.princeton.edu/~wn/). Choose five interesting pairs of words
and find sequences of WordNet relations that connect each word pair.
Miller
describes Wordnet as �an on-line lexical reference system whose design is
inspired by current psycholinguistic theories of human lexical memory. English
nouns, verbs, and adjectives are organized into synonym sets, each representing
one underlying lexical concept. Different relations link the synonym sets.�
We can
attribute various aims to the Wordnet enterprise. The two most important seem
to have been to a) harness modern computational power to create a more
convenient word catalogue that shared some of the properties of a thesaurus and
a dictionary, as well as incorporating new useful features of its own; b)
structure its relations so that they behave analogously to human lexical
memory, presumably in the hope that this would be useful for AI and NLP
systems.
Because
Miller considers �the starting point for lexical semantics � to be the mapping
between forms and meanings�, Wordnet incorporates both syntactic (e.g. part of
speech) and semantic (e.g. synonyms, is-a-kind-of, is-a-part-of and
is-one-way-to hierarchies) information. Further useful information is also
available, such as word commonality (measured in terms of polysemy), sentence
frames and dictionary definition glosses. The system allows you to traverse
effortlessly up, down and across the tree of all the words of the English
language structured by whichever relation you choose.
The paper
briefly discusses the representations used for nouns and their relations to
each other and words from other syntactic categories. Probably the most
interesting is their approach to broadly pinning down the �meaning� of a noun.
Although glosses are provided (and often sample sentences too), the aim of
Wordnet is not to try and provide enough information for a system to be able to
learn the meanings of all the words in the English language from a rudimentary
base. Rather, it is hoped that just enough information is provided to allow a
fluent English speaker to easy tell one word/sense from another.
Various
projects are underway to extend (freely-distributable and modifiable) Wordnet,
although only those sanctioned by the Princeton team are included with the
official �Wordnet� package. For example, a number of Euro Wordnet packages are
available that offer a similar systematisation of a number of languages besides
English, and the Extended Wordnet project aims to disambiguate and deconstruct
the glosses in terms of synsets.
I haven�t
used Wordnet for anything other than playing around, but I am impressed by it.
It appears fairly comprehensive, and it seems to me that if it didn�t already
exist, someone else would build it, since it provides a service that I would
think many AI and NLP projects would want to build up on at some stage. At the
very least, even if someone built a reasonably competent natural language
system which didn�t rely on Wordnet whatsoever, Wordnet would be of some use in
cementing its lexicon. Of course, it could be argued that any system in a
position to use Wordnet in such a way might well be close to being able to use
a normal dictionary. At this stage, I wouldn�t want to be pressed on exactly
how Wordnet might be useful to AI or NLP researchers.
Brian
mentioned that a handful of complaints are often levelled at it (namely that
there are too many disconnected synsets, the glosses in their present form are
useless, as well as the inevitable taxonomic disagreements), but broadly
speaking, I think it manages the tricky trade-off between machine- and
human-readability pretty well.
cat, true cat -- (feline mammal usually having thick soft fur and being
unable to roar; domestic cats; wildcats)
������ => feline, felid
���������� => carnivore
dog, domestic dog, Canis familiaris -- (a member of the genus Canis
(probably descended from the common wolf) that has been domesticated by man
since prehistoric times; occurs in many breeds; "the dog barked all
night")
������ => canine, canid
���������� => carnivore
cool (vs.
warm) -- (neither warm or very cold; giving relief from heat; "a cool
autumn day"; "a cool room"; "cool summer dresses";
"cool drinks"; "a cool breeze")
warm (vs.
cool) -- (having or producing a comfortable and agreeable degree of heat or imparting
or maintaining heat; "a warm body"; "a warm room"; "a
warm climate"; "a warm coat")
cool --
((informal) socially adept; "it's not cool to arrive at a party too
early")
antonym: INDIRECT (VIA felicitous) ->
infelicitous -- (not appropriate in application; defective; "an
infelicitous remark"; "infelicitous phrasing"; "the
infelicitous typesetting was due to illegible copy")
=> unfortunate (vs. fortunate) -- (not
favored by fortune; marked or accompanied by or resulting in ill fortune;
"an unfortunate turn of events"; "an unfortunate decision";
"unfortunate investments"; "an unfortunate night for all
concerned")
bad (vs.
good)
������ => atrocious, abominable, awful,
dreadful, painful, terrible, unspeakable
������ => corked, corky --
((of wine)
������ => deplorable,
distressing, lamentable, pitiful, sad, sorry
������ => fearful, frightful,
terrible
������ => fine, pretty
������ => hard, tough
������ => hopeless
������ => horrid
������ => icky, crappy, lousy,
rotten, shitty, stinking, stinky
������ => incompetent,
unskilled
������ => mediocre
������ => mischievous, naughty
������ => negative
������ => poor
������ => rubber, no-good
������ => severe
������ => swingeing
������ => unfavorable
������ => unsatisfactory
������ => unsuitable
������ => ill
I couldn�t
find an immediate link between �cool� and its synonyms/antonyms
(felicitous/infelicitous, fortunate/unfortunate) to �bad� and its antonyms
(good, favorable/unfavorable, satisfactory/unsatisfactory etc.) though I�m sure
there would be one within a couple more hops.
Also,
Wordnet did not contain �bad� as used in the informal sense to mean �really
good� (e.g. �that triple somersault was bad�) which was the sense I was
originally looking for.
hand,
manus, hook, mauler, mitt, paw -- (the (prehensile) extremity of the superior
limb; "he had the hands of a surgeon"; "he extended his
mitt")
��������� HAS PART: digital arteries, arteria
digitalis -- (arteries in the hand and foot that supply the fingers and toes)
��������� HAS PART: metacarpal artery, arteria
metacarpea -- (dorsal and palmar arteries of the hand)
��������� HAS PART: intercapitular vein, vena
intercapitalis -- (veins connecting the dorsal and palmar veins of the hand or
the dorsal and plantar veins of the foot)
��������� HAS PART: metacarpal vein, vena
metacarpus -- (dorsal and palmar veins of the hand)
��������� HAS PART: palm -- (the inner surface
of the hand from the wrist to the base of the fingers)
��������� HAS PART: finger -- (any of the
terminal members of the hand (sometimes excepting the thumb); "her fingers
were long and thin")
��������� HAS PART: ball -- (a more or less
rounded anatomical body or mass; ball of the human foot or ball at the base of
the thumb; "he stood on the balls of his feet")
��������� HAS PART: metacarpus -- (the part of
the hand between the carpus and phalanges)
�Scoff� is
a British idiomatic verb to mean �eat fast/greedily� (e.g. �he scoffed all the
cakes�). Wordnet did not list this as a sense of �scoff�.
1. jeer,
scoff, flout, barrack, gibe -- (laugh or scoff at; "The crowd jeered at
the speaker")
2. scoff,
flout -- (treat with contemptuous disregard; "flout the rules")