Linguistics for Language Technology

​

Week 6: Semantics

Lisa Bylinina

​

11 October 2023

Last week

• Grammar II: Syntax
• Infinite sentences using finite means
• Grammatical vs ungrammatical sentences
• Word order
• Morphosyntactic alignment: accusative vs ergative etc.

2

Today

• Semantics!

3

PART 1

INTRODUCTION TO SEMANTICS

​

• Sentence meaning
• Word meaning
• Their relation

4

What do you know when you know sentence meaning?

When we know the meaning of a sentence, we can distinguish situations which can be truthfully described by this sentence from situations in which it’s false

5

What do you know when you know sentence meaning?

• Truth conditions: Conditions under which the sentence is true
• Distinguish between truth and truth conditions
• We know ‘Paris is the capital of France’ is true in the real world…
• … but we also know what makes it true and how to check if this and other similar sentences are true (there’s a city called Paris, it’s in France, French government sits there etc. etc.)
• For some sentences we don’t know if they are true or false, but we are able to distinguish situations in which they are from those in which they are not
• When we know the meaning of a sentence, we know its truth conditions

6

Ways to think about truth conditions as sentence meaning

7

Truth conditions as sentence meaning

Potential problems:

• Do all sentences have truth conditions (can be true or false?)
• Only declarative ones!
• Some declarative sentences don’t easily fall into either category:

​

The king of France is bald.

8

What else do you know when you know sentence meaning?

• You know what you can conclude from a sentence
• = Which inferences you can draw and which inferences are not justified

​

A cat is sitting on a chair.

⇒ An animal is sitting on a chair.

⇒ A cat is sitting on a piece of furniture.

⇏ A cat is sitting on an old chair.

⇏ A cat is sitting on the floor.

9

Possible situations / possible worlds

Acciuga is a dog.

• We don’t know if this sentence is true because we don’t know Acciuga
• In order to know if it’s true, we would need to examine the situation with Acciuga
• As far as we are concerned now, we are facing a bunch of possible situations:
• It might be that Acciuga is a cat, a penguin, a professor, a mammal…

10

Possible situations / possible worlds

Whatever turns out to be true in the actual world, we know that:

• The set of situations in which Acciuga is a dog is a subset of the set of situations in which Acciuga is a mammal, so the latter follows from the former
• If we know that Acciuga is a mammal, we can’t conclude that A. is a cat or a dog – the Acciuga-mammal-situation set is a superset of Acciuga-dog/-cat

11

Possible situations / possible worlds

• {s | Acciuga is a dog in s} ⊆ {s | Acciuga is a mammal in s}
• {s | Acciuga is a mammal in s} ⊇ {s | Acciuga is a cat in s}
• {s | Acciuga is a cat in s} ∩ {s | Acciuga is a dog in s} = ∅

12

Truth conditions and possible situations

13

Direct vs. indirect interpretation

14

Towards word meanings

Zooming in to one possible state of affairs:

• There are some entities that are cats
• There are some entities that are mammals etc.
• The set of dogs is the subset of the set of mammals
• Sets of dogs and cats are disjoint…

15

From sentence meanings to word meanings

Connection between situations and entities:

• If Acciuga is in the set of dogs in the actual situation, then the actual situation is in the set of Acciuga-is-a-dog possible situations

16

Ways to think about noun meaning

17

Interpretation of proper names

18

Interpretation of other types of words

19

Interpretation of other types of words

20

Interpretation of other types of words

21

Semantic relations between words

22

Compositionality: meaning of whole from meaning of parts

23

Compositionality: meaning of whole from meaning of parts

24

Different structure, different meaning

25

Interpreting structures, not strings

26

PART 2

SEMANTICS AND LANGUAGE TECHNOLOGY

27

Semantics-related tasks

• The overall field of Natural Language Understanding:
• Summarization
• Question answering
• Information extraction
• …
• Word level:
• Word sense disambiguation
• Coreference resolution
• Sentence level:
• Semantic parsing
• Natural language inference

28

Lexical resources: WordNet

• https://wordnet.princeton.edu/
• Online search: http://wordnetweb.princeton.edu/perl/webwn
• Use in python:

29

Lexical resources: WordNet

30

Lexical resources: WordNet

31

Lexical resources: WordNet

32

33

34

35

36

Sentence-level resources: PMB (here in Groningen)

• https://pmb.let.rug.nl/

​

37

​

THANK YOU!

38