### Lecture 11 Outline

April 11, 2006

Assignment #9 ... discuss conditional probabilities

### Syntax-Driven Semantic Analysis  (J&M Chapter 15))

Complete oour discussion of semantic representations (Lecture 10 notes)

Review examples of logical forms

<>Discuss rules for producing semantic representations of sentences using syntax-driven semantic analysis
• principle of compositionality
(problem of idioms)
• basic idea:  nested S's and NP's
S's translate to predicates with arguments
NP's translate to complex terms
modifiers on NP's translate to predicates with one arg = quantified variable
• going all the way:  rule-by-rule (node-by-node) hypothesis
• associate a semantic feature (sem) with each node
• associate a semantic rule (in braces) with each node
• <><>problem of VP node:  lambda notation.

Scoping resolution:  converting quasi-logical form to logical form (with conventional quantifier scope)

Factors in resolving scope:

• ‘strength’ of different quantifiers
“each”, in particular, has wide scope
• order of quantifiers
(“Some students were rejected by every college.” vs. “Every college rejected some students.”)
• definite NPs evoking functional relationships
(“Every patient had an X-ray.”  vs. “An X-ray was taken of every patient.”)
• more generally, semantic knowledge
(“This train does not stop at every station.”)

### Strategies for data base query applications

Assuming queries are full grammatical sentences:
• Parse
• Convert to quasi-logical form
• Resolve scoping, generating logical form
• Interpret logical form as data base query:
• non-reified predicates access relations of data base
(different predicates may access different projections of same relation)
[in reified forms, event variables iterate over rows of data base]
• restricted quantifiers become iterations over sets of data base elements
•    [checking for non-empty sets:  presupposition violations]
• WH quantifiers print results
•    [quantifiers outside WH quantifier generate tables]
(we will consider how to handle fragmentary queries as part of discourse processing)

### A Canonical Representation?  (Lexical Semantics - J&M 16.1 - 16.2)

Logical form brings us a step closer to a 'canonical representation', where there is one representation for a meaning (and one meaning for a representation).  For example, active and passive forms will translate to the same logical form.

The predicates in our logical form are words, which is not satisfactory:
• words can have several senses ("Fred beat Mary.").  Resolving these ambiguities is the task of word sense disambiguation (J&M 17.1-17.2)  The result would be a logical form in which the predicates are word senses.
• several word senses may have the same meaning ("stop", "halt").  We can group word senses into synonym sets ("synsets"), and use synsets as predicates.  WordNet provides a large inventory of English synsets (J&M sec. 16.2), and there are WordNets now for many languages (see the Global WordNet Association).
But this far from exhausts the range of paraphrases (different ways of conveying the same meaning).  Building larger inventories of paraphrases remains a very difficult problem for NLP.