Title: Computational Linguistics
1Computational Linguistics
2Introduction
- Computational Linguistics within Artificial
Intelligence - Linguistics what it is, past influences
- Language underlying set of rules
- Linguistics todays issues and how they impact
computational linguistics
3Computational Linguistics
- sub-field of Artificial Intelligence
- Natural Language Processing (NLP)
- used in interfaces, search engines etc.
- applies techniques developed in Linguistics
4Artificial Intelligence
- Connectionism
- based roughly on our brain neural nets
- nodes in network
- network learns - connections strengthen
- Classical AI
- symbols plus rules for manipulating symbols
- NLP
- words symbols
- grammar rules rules to manipulate words.
5Classical AI
- successful in limited domain
- (eg Big Blue chess - used brute force)
- not so good when problems imbedded in world
- NLP
- When she hit the nail with the hammer the head
flew off. - requires world knowledge to disambiguate
6Connectionism
- node (computing unit)
- connections between nodes vary in strength
- learning rule error signal ideal - actual
- input output error signal
7Connectionism..some success
- connectionist net was trained to learn the past
tense of verbs. (Rummelhart McClelland 1986) - followed same U shaped learning pattern of
children - went
- learn rule over generalized goed
- master both regular and irregular past tense
8Connectionismnot solution
- when children hear a sentence
- The dog chased the cat.
- spontaneously treat cat as if belonging to the
same category as dog - neural nets fail to generalize in this way...
9Unique Challenges
- not all problems similar to other areas of AI
- for example early search engines
- what is a word?
- computer equals computers but not computation
- need a deeper analysis of language linguistics
10Linguistics
- history
- evolutionary history of sets of related languages
- structures
- the structures that make up language
- (SVO-English) (SOV-Japanese) (VSO-Arabic)
- use
- what do words and sentences mean? Mental
representations? Function of use?
11Sub-fields of Linguistics
- phonetics (sounds in language)
- where sounds are made
- phonology (sound patterns in language)
- what sound combinations are allowed
- morphology (words and word structures)
- rules for combining morphemes
- syntax (sentence and sentence structures)
- rules for coming words in a sentence
- semantics (meaning)
12Grammar
- descriptive
- how languages are actually used
- prescriptive
- the correct way to speak, based on current
standards - theoretical
- theories of language structures
- formalize rules we unconsciously apply when
speaking
13Examples of Rules We Apply(From English)
- Morphology
- encounter new verb in present tense flig
- would use it in the past tense fligged
- Phonology
- ptil, sorlint, bkat
- which one could be an English word?
14Unconscious Rules
- only sorlint is a candidate for a new English
word because - Phonological Rule in English
- words cannot begin with two stop consonants
- p, b, t, d, k, g
15Linguistics Early 20th Century
- structuralism classifying/taxanomic
- Focus on language form (S -gt NP VP) as opposed to
meaning (S -gt complete thought) - Edward Sapir
- analyzed Amerindian Languages
- Bloomfield
- influenced by behaviourism
- WWII
- need to analyze and learn many new languages
16Language Form Underlying System of Rules
- Phrase Structure Rules
- Representing sentences tree structures
- Parts of Speech
- Constituents and Categories
- Ambiguity more than one possible way to
parse sentence
17Phrase Structure Rules
- Rules Lexicon
- S NP VP Det the
- NP Det N N child
- VP V V sleeps
18Representation
- tree structures used to represent sentences
- sentences broken down into constituents
- S
- NP VP
- Det N V
- The child sleeps
19Parts of Speech
- NOUNS
- person, animal, place, quality, idea
- Mary, university, thought, childhood
- PRONOUN
- refer to, or stand in for, a noun
- everyone, she, they
- VERB
- action, state of being
- walk, to be (is, are), know
20Parts of Speech
- ADJECTIVE
- describes or modifies a noun
- happy, red, many (Many people like chocolate)
- ADVERB
- modifies a verb, adjective, adverb, phrase,
clause, sentence - slowly, very slowly, immediately (Immediately,
they...) - PREPOSITION
- links nouns, pronouns, and phrases to other parts
of the sentence - express relationships in time
or space between things and events - to, before, beside, on, at, of, under, since,
after.
21Parts of Speech
- CONJUNCTION
- joins two or more words, phrases, or clauses
- and, but, or
- INTERJECTION
- used to protest, exclaim, command
- oh, wow, hey
- DETERMINER
- articles the, a/an (only two in English)
- demonstratives (modifies noun, indicates
position of something in relation to speaker)
that, those, this, these
22Constituents and Categories
- Categories parts of speech
- (nouns, verbs, etc.)
- Constituents words that hang together
- can move together
- can be deleted together
- makes sense to talk of the meaning of a
constituent structure (over other combinations)
23Constituents and Categories
- S
- NP VP
- det N V NP PP
- det N P NP
- det N
- the dog chased the cat into the
garden
24Ambiguity in Phrase Structure
- ambiguity can result when there is more than one
possible phrase structure for a sentence - Example
- The girl saw the boy with the telescope.
- Who has the telescope?
- The girl? Or the boy?
25Phrase Structure Rules
- S NP VP det the
- NP det N N girl, boy, telescope
- NP det N PP P with
- PP P NP V saw
- VP V
- VP V NP
- VP V NP PP
-
26Phrase Structure Rules
- The girl has the telescope.
- S
- NP VP
- det N V NP PP
- the girl saw
- det N P NP
- the boy
with -
det N - the telescope
27Phrase Structure Rules
- The boy has the telescope.
- S
- NP VP
- det N V NP
- the girl saw
- D N PP
- the boy
- P NP
- with
- D N
-
the telescope
28Phrase Structure Rules Complete Grammar?
- no
- many constructions impossible to describe
adequately
29Wh movement
- Jane said Paul believed Fido barked.
- Who said Paul believed Fido barked? (Jane)
- Who did Jane say believed Fido barked? (Paul)
- Who did Jane say Paul believed barked? (Fido)
30Transformational Rules
- Wh - movement any amount of structure between
original position of moved word and where it ends
up. - Phrase structure rules cannot account for it.
31Transformational-Generative Grammar
- Noam Chomsky
- Syntactic Structures (1957)
- 1 - 100 most influential works in Cognitive
Science in the 20th Century - A Review of Skinners Verbal Behaviour (1959)
- Together launched Cognitive Revolution
- mind rules representation
- descriptive theories to explanatory theories
32Universal Grammar
- language is innate
- each language is an example of UG
- biological basis of language
33Universal Grammar Requires
- descriptive adequacy
- detailed enough to describe any possible
construction in any human language - explanatory adequacy
- simple enough to reflect the small set of inborn
principles that allow humans to acquire and use
language
34Tension between two requirements of UG
- A theory has explanatory adequacy to the extent
that it is able to show how, from the data
available to a child, a child can get a
descriptively adequate grammar. - to be more descriptively adequate, need more
descriptive devices - more devices, harder it is for a child to gain
descriptively adequate grammar from data
35How does this help computational linguistics?
- Phrase structure rules inadequate to describe
language. - Transformational rules difficult to parse -
seldom used in NLP. - TG caused explosion in rules.
- Chomsky has since moved on to minimalist
program.
36Minimalist Program
- two levels of representation
- phonetic form
- logical form
- implication of language being innate - language
has evolved (not mathematically elegant) - does not follow rules perfectlyin some respects
unusable
37Is natural language processing possible?
- One reason for studying language - and for me
personally the most compelling reason - is that
it is tempting to regard language, in the
traditional phrase, as a mirror of mind. - Chomsky, 1975