One way of approaching linguistics is to describe generalities at each level.
Another way of approaching it is to say that knowing a language is knowing how to use its words.
Each word has:
Semantic aspect
Syntactic aspect
Collocational aspect
Available at: www.cogsci.princeton.edu/~wn/.
Deals only with content words.
Imposes a matrix on the lexicon (word forms vs. meanings).
Maps 57K word forms onto 49K synonym sets ('synsets')
Works largely by declaring links among words synonym sets.
Result is a hierarchical network. (Trees).
Computers facilitate browsing the network. Hard copy dictionaries don't.
[ Up to An example of a lexical database: Wordnet]
Wordnet bases its organization on studies of psycholexicology.
Aphasias (language disabilities due to brain damage) give hints at the relationship between various linguistic abilities and brain structure. E.G: Some people lose only the ability to name things.
Differences in response time associating words can be taken as evidence of a network structure associationg various terms, with longer reponse times possibly reflecting a traversal of more links in the network.
[ Up to An example of a lexical database: Wordnet]
Polysemy is ambiguity of meaning.
Homonyms
Synonymy: words with similar meanings
Words that can be used interchangably in all instances are rare. (Ever hear of a 'sofa potato?')
Whether two words go together often largely a matter of degree.
[ Up to An example of a lexical database: Wordnet]
Not the same as 'Not ~'
Lexical: Applies between word forms rather than meanings.
EG: rise/fall, ascend/descend
[ Up to An example of a lexical database: Wordnet]
[ Up to Nouns]
Actions
Animals
Artifacts
Attributes
Body
Cognition
Communication
Events
Feelings
Food
Groups
Locations
Motives
Natural Objects
Natural phenomena
Persons
Plants
Possessions
Processes
Quantities
Relations
Shapes
States
Substances
Time
[ Up to Nouns]
Based on 'is-a' relations.
eg: Red -> color.
eg: Lion -> feline.
Typical dictionary uses 'superordinate + distinguishers'. EG: A tree is a plant which is tall and has a trunk.
Attributes of parents acrue to descendents.
Inheritance is Transitive and asymetrical. 'Transitive means that if an oak is a tree and a tree is a plant, then an oak is a plant. Asymetrical means that if an oak is a tree then a tree is not necessarily an oak.
Problems
[ Up to Nouns]
Has-a/part-of relation
EG: tire->wheel->car
EG: ?handle->door->house
6 types *(3 used in wordnet)
*Component-object (branch/tree)
*Member-collection (tree/forrest)
Portion-mass (slice/cake)
*Stuff-object (aluminum/airplane)
Feature-activity(paying/shopping)
Place-area (San Diego/California)
[ Up to Nouns]
Uses metaphorical extension.
EG: 'Washington'
EG: 'The Brass'
It is common in some linguistic circles to use metaphor to refer only to anomalous cases. There are other schools of thought which claim that metaphor actually pervades all of language to some degree (cf George Lakoff's Women, Fire, and Dangerous Things).
[ Up to Nouns]
Which nouns typically participate in which verbs?
This would be quite desirable from an NLP perspective, but probably hard to define. 'Eat' may involve an animal and food, but cows do not eat meat.
Not implemented in wordnet.
[ Up to An example of a lexical database: Wordnet]
Adjectives modify nouns.
They operate on a domain provided by the noun head.
Prenomial/Postnomial/Predicative. 'Red shoe'/'president elect'/'The patient is asleep' (but not *'the asleep patient'), repectively.
Descriptive adjectives pertain to the attributes that pertain to nouns
Bipolar
EG: 'Heavy'
Relational adjectives ('of or pertaining to') are not bipolar.
EG: 'musical'
Some adjectives apply to domains which vary gradually, while others are 'hit or miss'.
Not all adjectives have antonyms, eg colors.
[ Up to An example of a lexical database: Wordnet]
More nouns than verbs (43K vs 14K)
Hierarchies are much shallower.
Verbs are more mutable: polysemy=2.11 vs 1.74
Wordnet does not do semantic decomposition.
[ Up to Verbs]
Verbs can be characterized on the basis of what arguments they take and how they take those arguments.
EG: John saw Mary
EG: John gave Mary an apple.
EG: John gave an apple to Mary.
EG: John made a pot out of clay.
EG: Mary made clay into a pot.
EG: *John fabricated clay into a pot.
EG: Mary said the apple is good.
EG: Mary believes the apple is good
EG: Mary believes the apple to be good.
EG: *Mary said the apple to be good.
[ Up to Verbs]
Rise/fall applies to temperature.
Ascend/descend does not.
As you move down the hierarchy, selectional restrictions tighten.(eg:dine->eat->consume)
[ Up to Verbs]
Bodily functions
Change
Communication
Competition
Consumption
Contact
Cognition
Creation
Motion
Emotion
Stative
Perception
Possession
Social interaction
Weather
[ Up to Verbs]
The verbal equivalent of meronymy.
+/- 'temporal inclusion'
EG: snore->sleep
[ Up to Verbs]
Verbal equivalent of hyponymy
Is a manner of ...
EG: limp -> walk
Co-extensive (in contrast to 'snoring')
[ Up to Verbs]
EG: give->have; buy->own
Usually [action] ...[state]
[ Up to Verbs]
EG:untie->tie
EG:return->depart
[ Up to An example of a lexical database: Wordnet]
Disambiguation: consulting a larger context than that provided by N-grams and finding words which are associated with only one sense of a word may provide valuable clues about which word sense is being used in a particular context.
Many information retrieval systems for free text rely on statistical matching between the words in the query and those in documents in the database. On way of increasing the recall of relevant documents is to use a thesaurus. A lexical database like wordnet can extend that further by identifying words which are likely to be found in the same context as words relevant to the query.