Language and Communication

1
Language and Communication

• Sentence Processing

2
What do we do when we Process a Sentence?

• Perceptual processes recognise words in a
  sentence.
• ? lexical access provides syntactic and semantic
  knowledge to determine what kind of word each
  word is, and what it means
• ? parsing/interpretation syntactic and semantic
  combination rules are used to put the words
  together into meaningful groups (and to group the
  groups) and to determine the literal meaning
• ? pragmatics determines the meaning intended in
  the context

3
What is the Relation between the Stages of
Processing?

• Two views
• Lexical access and syntactic parsing are
  autonomous (self-contained modules) modular
  view.
• Semantic (and contextual) information can be used
  to guide lexical access and syntactic analysis
  interactive view.

4
Does information flow back down the dotted lines?
Modular view says no
Interactive view says yes
From Whitney, pp. 205.
5
Ambiguity

• Lexical ambiguity
• A sentence contains a word with more than one
  meaning, and the appropriate meaning is usually
  determined from the context (see previous
  lecture).
• Syntactic ambiguity
• Where a clause or sentence may have more than
  one interpretation given the potential
  grammatical functions of the individual words
  (Berko-Gleason, pp.238)
• The processing of ambiguous sentences turns out
  to be a crucial test case for theories of
  syntactic processing

6
Syntax

• The words in a sentence are not just strung
  together one after the other, they are structured
  into phrases and clauses
• Typical phrases might be the man (a noun
  phrase) and saw the man (a verb phrase)
• Note that one phrase can occur within another
• The way that words are structured into phrases,
  clauses and sentences can be depicted in tree
  diagrams
• The trees are upside down with the root at the
  top
• In understanding a sentence, we need to establish
  the groupings of the words so that we can go on
  to work out what the groups, and hence what the
  sentence, means.

7
Phrase (or Tree) Structure Rules

• We can write rules that show how to build tree
  structures (or we can think of them as rules for
  analysing sentences)
• S ? NP VP
• (a sentence can be made up of a noun phrase
  followed by a verb phrase)
• E.g. Jim slept OR The boy hit the ball
• NP ? DET N
• (a noun phrase can be a determiner an article
  or similar followed by a noun
• E.g. The boy OR A ship
• NP ? N
• (a noun phrase can be just a noun e.g. Jim OR
  ships)
• VP ? V NP
• VP ? V
• (a verb phrase can be a verb followed by a noun
  phrase or it can be just a verb)

8
Phrase (or Tree) Structure Rules

• Then we have rules to introduce individual words
  at the bottom of the (upside down) trees
• V ? hit, meet, jump..
• DET ? the, a, this, that.
• N ? boy, ball, Jim,..

9
Syntactic Tree Structures
S
The person doing something
The thing being done
NOUN PHRASE
VERB PHRASE
The thing that was having something done to it
NOUN PHRASE
DET
N
V
ADJ
N
DET
THE BOY HIT THE RED BALL
SUBJECT
OBJECT
VERB
10
Additional rules

• VP ? V PP
• PP ? P NP
• PP Prepositional phrase, P preposition

11
THE YOUNG MAN STOOD IN THE HOLE
12
Some examples to parse..

• The puppy found the child.
• The house on the hill collapsed in the wind.
• The ice melted.
• The hot sun melted the ice.
• A quaint old wooden house appeared.
• The old tree swayed in the wind.

13
Reminder

• Syntactic ambiguity
• Where a clause or sentence may have more than
  one interpretation given the potential
  grammatical functions of the individual words
  (Gleason, pp.238).
• Local ambiguity a sentence is only temporarily
  ambiguous.
• After he rang his mother called the police
• Global ambiguity sentence remains ambiguous
  after all the words have been considered.
• The cop saw the man with the binoculars

14
Sentence Processing - Syntactic Ambiguity

• Garden path sentences
• In which the early part of the sentence make the
  reader or listener goes down the garden path
  and get the wrong interpretation
• The horse raced past the barn fell (Bever, 1970)
• The old man the boats.
• Since Jay always jogs a mile seems like a short
  distance (Frazier Rayner, 1982).
• Why do we prefer one interpretation of a
  syntactically ambiguous sentence to another?
• What is the role of context in resolving
  syntactic ambiguity? Modular versus interactive
  views.

15
Real Headlines (Global Ambiguities)

• Red tape holds up new bridge
• Retired priest may marry Springsteen
• British left waffles on Falklands
• Safety Experts Say School Bus Passengers Should
  Be Belted
• Panda Mating Fails Veterinarian Takes Over
• Squad Helps Dog Bite Victim

16
Real Headlines

• Miners Refuse to Work after Death
• Juvenile Court to Try Shooting Defendant
• Kids Make Nutritious Snacks
• Local High School Dropouts Cut in Half
• New Vaccine May Contain Rabies
• Hospitals are Sued by 7 Foot Doctors
• Westinghouse gives robot rights to firm

17
Church Notices

• Don't let worry kill you let the church help.
• For those of you who have children and don't know
  it, we have a nursery downstairs.
• The ladies of the church have cast off clothing
  of every kind. They can be seen in the church
  basement Saturday.

18
The Problem

• How do we characterise
• the different kinds of AMBIGUITY?
• the preference for one reading over another?
• Phrase structure trees provide a way of thinking
  about possible explanations

19
Two Models

• Garden path model only access one meaning - late
  closure and minimal attachment.
• Constraint satisfaction model access both
  possible meanings but are conscious of only one.

20
The Garden Path TheoryLyn Frazier, 1970s and
1980s

• Build trees using phrase structure rules, plus
  two principles that are also syntactic in nature
  (i.e. they do not refer to meaning)
• Minimal Attachment
• Prefer the interpretation that is accompanied by
  the simplest structure.
• simplest fewest branchings (tree metaphor!)
• Count the number of nodes branching points
• Late Closure
• Incorporate incoming material into the phrase or
  clause currently being processed. OR
• Associate incoming material with the most recent
  material possible.

21
Parsing preferences minimal attachment
S
8 Nodes Preferred
NP
VP
the girl
V
NP
S
hit
NP
PP
NP
VP
the man
P
NP
the girl
V
NP
PP
with
the umbrella
hit
the man
P
NP
with
the umbrella
9 nodes Dispreferred
The girl hit the man with the umbrella.
22
Parsing Preferences .. late closure
S
Preferred
S
np
vp
np
vp
she
v
S'
adv
she
v
S'
said
np
vp
yesterday
said
np
vp
he
v
np
he
v
np
adv
tickled
her
tickled
her
yesterday
Dispreferred (Both have 10 nodes, so use LC not
MA)
She said he tickled her yesterday
23
S
NP
VP
S
the spy
V
PP
NP
saw
P
NP
the cop
with
the binoculars
but the cop didnt see him
The spy saw the cop with the binoculars but the
cop didnt see him.
24
S
NP
VP
the spy
V
NP
S
saw
NP
PP
the cop
P
NP
with
the binoculars
but the cop didnt see him
25
(No Transcript)
26
Non-MA
MA
The spy saw the cop with the revolver but the cop
didnt see him.
27
Constraint Satisfaction ModelTrueswell, et al.,
1994, and MacDonald, 1993

• Different syntactic structures are activated by
  presence of words, word types (NOUN, VERB etc.)
  or sequences of words or word types
• e.g. a DET followed by a NOUN will activate NP ?
  DET N syntactic structure
• Strength of activation is dependent on the
  frequency of occurrence.
• E.g. complement clauses are more common than
  relative clauses, so ..told the woman that..
  activates VP ? V NP S more strongly than NP ? NP
  RelClause

28
Constraint Satisfaction Model - cont

• Other factors, such as semantic context, also
  provide constraints on the likely interpretation
  of a sentence, and may allow some syntactic
  structures to be ruled out very early during
  syntactic analysis
• e.g. The evidence questioned .
• The person questioned .
• The model leaves it open how quickly the
  different kinds of constraint can act.
• If semantic constraints act quickly, the model is
  interactive
• Whether modular or interactive view is correct is
  still an issue.

29
Semantic expectations

• Taraban McCelland, 1988, semantic expectations
  mean that when we read The spy saw.. this leads
  us to expect that the remainder of the sentence
  will describe what the spy used and it is this
  expectation, rather than MA, that leads to
  momentary confusion when see the noun revolver
  as this could not be used by the spy.
• The couple admired the house with a friend but
  knew that it was over-priced.
• The couple admired the house with a garden but
  knew that it was over-priced.

30
Taraban and McClelland - resultsThe Non-MA
structure may be favoured
Taraban McCelland, 1988 , J of memory and
language, 27, 597-632.
31
Another Minimal Attachment Ambiguity?

• He told the woman that he was worried about to
  wait outside.
• her.

32
Structures ...
8 nodes Dispreferred
6 nodes Preferred
33
Context Effects

• Is syntactic ambiguity resolved only on the basis
  of syntactic information?
• Alternative syntactic analyses alternative
  meanings...
• Could syntactic ambiguity be resolved on the
  basis of meaning?
• Using a relative clause presupposes that one
  needs to identify the thing being referred to
• e.g. the woman that he was worried about

34
Referential Context

• Altmann et al. (1992)
• complement clause
• He told the woman that he was worried about many
  other people
• relative clause
• He told the woman that he was worried about to
  wait outside
• control
• He asked the woman that he was worried about to
  wait outside

35
Eye-Tracking Data (Reading Times)
36
Factors Implicated in Ambiguity Resolution

• structural
• referential context
• temporal context
• the horse raced past the barn fell
• any horse raced past the barn will fall
• real world knowledge
• after the police stopped the driver fled
• after the lorry stopped the driver fled
• relative frequency of usage of different meanings
  and/or structures.

37
Conclusion.. Questions remain

• Is syntactic ambiguity resolved only on the basis
  of syntactic information?
• Could syntactic ambiguity be resolved on the
  basis of meaning?
• Issue autonomous versus interactive model?