Transcript Natural Language Understanding

Natural Language Processing
(3)
Zhao Hai 赵海
Department of Computer Science and Engineering
Shanghai Jiao Tong University
[email protected]
1
Outline

Lexicons and Lexical Analysis

Generative Lexicon (1)
2
Lexicons and Lexical Analysis (31)
Generative Lexicon (1)
Overview

This lexicon deals with a novel and exciting theory of lexical semantics
that addresses the problem of the “multiplicity of word meaning”, that is,

how we are able to give an infinite number of senses to words with finite
means.

The inventor, James Pustejovsk, proposed that the lexicon becomes an
active and central component in the linguistic description.
3
Lexicons and Lexical Analysis (32)
Generative Lexicon (2)
Functions of the Lexicon
(linguistically motivated only)
The essence of his theory is that the lexicon functions generatively,
 by providing a rich and expressive vocabulary for characterizing
lexical information;
 by developing a framework for manipulating fine-grained
distinctions in word descriptions;
 by formalizing a set of mechanisms for specialized composition
of aspects of such descriptions of words, as they occur in context,
extended and novel senses are generated.
4
Lexicons and Lexical Analysis (33)
Generative Lexicon (3)
Problems on Lexical Semantics
Formal theories of natural language semantics have done little to
explore two important issues:

The creative use of words in novel contexts;

An evaluation of lexical semantic models on the basis of
compositionality.
5
Lexicons and Lexical Analysis (34)
Generative Lexicon (4)
Computational Lexical Semantics

a clear notion of semantic well-formedness will be
necessary in order to characterize a theory of possible word
meaning;

lexical semantics must look for representations that are richer
than thematic role descriptions;

lexical semantics must study all syntactic categories in order
to characterize the semantics of natural language.
6
Lexicons and Lexical Analysis (35)
Generative Lexicon (5)
Sense Enumeration Lexicon (SEL)
A lexicon L is a SEL if and only if for every word w in L, having
multiple senses s1, s2, …, sn associated with that word, then the
lexical entries expressing these senses are stored as
{ws1, ws2, …, wsn}.
That means it is to allow the lexicon to have multiple listings of
words, each annotated with a separate meaning or lexical sense.
7
Lexicons and Lexical Analysis (36)
Generative Lexicon (6)
Lexical Descriptions (Nouns)
bank1
CAT = count_noun
GENUS = financial_institution
bank2
CAT = count_noun
GENUS = shore
8
Lexicons and Lexical Analysis (37)
Generative Lexicon (7)
Lexical Descriptions (Verbs)
The bank will lend the money to the customer.
lend1
CAT = verb
SEM = R0(1, 2, 3)
ARGSTR =
ARG1 = np [+financial_institution]
ARG2 = np [+money]
ARG3 = np [+human]
9
Lexicons and Lexical Analysis (38)
Generative Lexicon (8)
Contrastive Ambiguity [Weinreich, 1964] (1)
A lexical item accidentally carries two distinct and unrelated
meanings.
Ex. 1:
a) Drop me a line when you are in Boston. (a short personal letter)
b) We built a fence along the property line. (a boundary)
10
Lexicons and Lexical Analysis (39)
Generative Lexicon (9)
Contrastive Ambiguity [Weinreich, 1964] (2)
Ex. 2:
a) The discussion turned on the feasibility of the scheme.
(depend upon)
b) The bull turned on the matador. (become hostile towards)
11
Lexicons and Lexical Analysis (40)
Generative Lexicon (10)
Complementary Polysemies [Weinreich, 1964] (1)
The lexical senses clearly show the same basic meaning of a
word as it occurs in different contexts.
Ex. 1:
a) The bank raised its interest rates yesterday.
(an institution)
b) The store is next to the newly constructed bank.
(a building)
12
Lexicons and Lexical Analysis (41)
Generative Lexicon (11)
Complementary Polysemies [Weinreich, 1964] (2)
Ex. 2:
a) The farm will fail unless we receive the subsidy promised.
(workplace)
b) To farm this land would be both foolish and without
reward. (working mode)
13
Lexicons and Lexical Analysis (42)
Generative Lexicon (12)
Complementary Polysemies [Weinreich, 1964] (3)
Note: The logical polysemy denotes complementary ambiguity
where there is no change in lexical category, and the multiple
senses of the word have overlapping, dependent, or shared
meanings.
14
Lexicons and Lexical Analysis (43)
Generative Lexicon (13)
New Definition for SEL
A lexicon L is a SEL if and only if for every word w in L,
having multiple senses s1, s2, …, sn associated with that word,
then:
a) if s1, s2, …, sn are contrastive senses, the lexical entries
expressing these senses are stored as ws1, ws2, …, wsn.
b) if s1, s2, …, sn are complementary senses, the lexical entry
expressing these senses is stored as w{s1, s2, …, sn}.
15
Lexicons and Lexical Analysis (44)
Generative Lexicon (14)
Complementary Senses
Ex.
a) John ate lamb for breakfast.
b) The lamb is running in the field.
lamb
SENSE1 = CAT = mass_noun
GENUS = meat
SENSE2 = CAT = count_noun
GENUS = animal
16
Lexicons and Lexical Analysis (45)
Generative Lexicon (15)
Limitations of SEL



The creative use of words: Words assume new senses in
novel contexts;
The permeability of word senses: Word senses are not
atomic definitions but overlap and make reference to other
senses of the word;
The expression of multiple syntactic forms: A single word
sense can have multiple syntactic realization.
17
Lexicons and Lexical Analysis (46)
Generative Lexicon (16)
Creative Use of Words (1)
Ex.
1. The island authorities sent out a fast little government boat, the Culpeper,
to welcome us: a boat driven quickly or a boat that is inherently fast
2.
3.
a fast typist: a person who performs the act of typing quickly
Rackets is a fast game: the motions involved in the game are rapid and
swift
4.
a fast book: one that can be read in a short time
18
Lexicons and Lexical Analysis (47)
Generative Lexicon (17)
Creative Use of Words (2)
fast1 to move quickly
Ex. 1
fast2 to perform some acts quickly
Ex. 2, 3
fast3 to do something that takes little time
Ex. 4
19
Lexicons and Lexical Analysis (48)
Generative Lexicon (18)
Creative Use of Words (3)
Ex.
5.
6.
The Autobahn is the fastest motorway in Germany.
I need a fast garage for my car, since we leave on Saturday.
Ex. 5: fast4: the ability of vehicles on the motorway to sustain
high speed (new sense for novel context)
Ex. 6: fast2 + fast3: the length of time needed for a repair by the
garage
20
Lexicons and Lexical Analysis (49)
Generative Lexicon (19)
Creative Use of Words (4)
Conclusion:
Enumeration is unable to exhaustively list the senses that
adjectives or verbs assume in new context!
21
Lexicons and Lexical Analysis (50)
Generative Lexicon (20)
Permeability of Word Senses (1)
Even if we were to assume that sense enumeration were
adequate as a descriptive mechanism, it is not always obvious
how to select the correct word sense in any given context.
Ex.
1. John baked the potatoes. (change-of-state)
2. Mary baked a cake. (creation)
22
Lexicons and Lexical Analysis (51)
Generative Lexicon (21)
Permeability of Word Senses (2)
Ex.
1. Mary cooked a meal.
2. Mary cooked the carrots.
3. John fried an omelet.
4. John fried an egg.
23
Lexicons and Lexical Analysis (52)
Generative Lexicon (22)
Permeability of Word Senses (3)
Problems:
1. There is too much overlap in the core semantic components
of the different readings;
2. It lacks any appropriate or natural level of abstraction.
24
Lexicons and Lexical Analysis (53)
Generative Lexicon (23)
Expression of Multiple Syntactic Forms (1)
The following sentences show that the syntactic realization of
the verb’s complement determines how the proposition is
interpreted semantically.
1. Madison Avenue is apt to forget that most folks aren’t
members of the leisure class. (factive)
25
Lexicons and Lexical Analysis (54)
Generative Lexicon (24)
Expression of Multiple Syntactic Forms (2)
2.
But like many others who have made the same choice, he
forgot to factor one thing into his plans: Caliphobia. (non-
factive)
3.
What about friends who forget the password or never got it?
(concealed question)
26
Lexicons and Lexical Analysis (55)
Generative Lexicon (25)
Expression of Multiple Syntactic Forms (3)
forget1
CAT = verb
SEM = R2(1, 2[-FACTIVE])
ARGSTR = AGR1 = np
AGR2 = vp[+inf]
forget2
CAT = verb
SEM = R2(1, 2[+FACTIVE])
ARGSTR = AGR1 = np
AGR2 = s[+tns]
27
Lexicons and Lexical Analysis (56)
Generative Lexicon (26)
Expression of Multiple Syntactic Forms (4)
forget3
CAT = verb
SEM = R3(1, 2)
ARGSTR = AGR1 = np
AGR2 = np
The proper expression is to have one definition for “forget”
which could generate all the allowable readings by suitable
composition with the different complement types.
28
Lexicons and Lexical Analysis (57)
Generative Lexicon (27)
Expression of Multiple Syntactic Forms (5)
The range of subjects possible with causative and experiencer
verbs:
Ex.
1. Driving a car in Shanghai frightens me.
2. Driving frightens me.
3. John’s driving frightens me.
4. Cars frightens me.
29
Lexicons and Lexical Analysis (58)
Generative Lexicon (28)
Expression of Multiple Syntactic Forms (6)
The syntactic argument to a verb is not always the same logical
argument in the semantic relation.
30
Lexicons and Lexical Analysis (59)
Generative Lexicon (29)
Description Languages for Lexical Semantics (1)
1. Monomorphic Languages (MLs)

The lexical items and complex phrases are provided a single
type and denotation.

Lexical ambiguity is handled by having a multiple listing of
the word.
Such as the description language used in SEL.
31
Lexicons and Lexical Analysis (60)
Generative Lexicon (30)
Description Languages for Lexical Semantics (2)
2. Unrestricted Polymorphic Languages (UPLs)

No restriction on the type that a lexical item may assume.

No operational distinction between subclasses of
polymorphic transformations.
For example, the theory of Searle (1979).
32
Lexicons and Lexical Analysis (61)
Generative Lexicon (31)
Description Languages for Lexical Semantics (3)
3. Weakly Polymorphic Languages (WPLs)

All lexical items are semantically active, and have a richer
typed semantic representation than conventionally assumed;

Semantic operations of lexically-determined type changing
operate under well-defined constraints;
33
Lexicons and Lexical Analysis (62)
Generative Lexicon (32)
Description Languages for Lexical Semantics (4)

Different subclasses of polymorphic operations are defined,
each with independent properties and conditions on their
applications.
34
Lexicons and Lexical Analysis (63)
Generative Lexicon (33)
Description Languages for Lexical Semantics (5)
What natural language data seem to require is a semantic system
falling outside of ML, but well below the language of UPL:
ML  WPL  UPL
It is important to look more at the generative or compositional
aspects of lexical semantics, rather than decomposition into a
specified number of primitives (senses).
35
Lexicons and Lexical Analysis (64)
Generative Lexicon (34)
References (1)
J. Pustejovsky. 1989. Issues in Computational Lexical
Semantics. In the Proc. of the Fourth European ACL Conference,
Manchester, England.
J. Pustejovsky. 1991. The Generative Lexicon. Computational
Linguistics, Vol. 17, pages 409-441.
J. Pustejovsky & B. Boguraev. 1993. Lexical Knowledge
Representation and Natural Language Processing. Artificial
Intelligence, Vol. 63, pages 193-223.
36
Lexicons and Lexical Analysis (65)
Generative Lexicon (35)
References (2)
J. Pustejovsky & B. Boguraev. 1995. Lexical Semantics in
Context. Journal of Semantics, Vol. 12, pages 1-14.
37