Conflicts in Interpretation

1
Conflicts in Interpretation

• Henriëtte de Swart
• UiL-OTS/Utrecht

2
Polysemy of round

• The postman ran around the block

• Prototype of round circle, circular shape or
  movement

3
Less prototypical I

• He drove round the barrier.

• He ran round the corner.

4
Less prototypical II

• The captain sailed round the lake

• The tourist drove round the city centre.

5
Round

• Interpretation interaction of lexical semantics
  (cognitive grammar), model-theoretic semantics
  (spatial path), and pragmatics (world knowledge).
• Consequently cross-modular approach.
• Conflicting information
• Differentiated weight

6
Interpretation as Optimization

• Optimization in neural network modeling
• Application to theories of grammar Prince and
  Smolensky (1991, 1997).
• Optimality Theory
• Phonology
• Syntax
• Semantics Hendriks and de Hoop (2001), de Hoop
  and de Swart (2000)

7
OT Semantics

• Input form.
• Output candidate set of possible
  interpretations.
• Parallel evaluation
• Violable constraints.
• Ranking determines strength.
• Optimal interpretation winner takes all.

8
Conflicts in Interpretation

• Cognition Program
• 2002-2006
• Research teamPetra Hendriks, Gerlof Bouma
  (Groningen), Helen de Hoop, Irene Krämer
  (Nijmegen), Henriëtte de Swart, Joost Zwarts
  (Utrecht)

9
Research Themes

• Cross-modularity (focus, pronoun resolution)
• Cross-linguistic semantics (bare nominals,
  lexical semantics, negation)
• Language Acquisition (acquisition of anaphoric
  relations, noun phrases).
• Central concern role of context.

10
Interpretation in context

• Semantics of round Zwarts (2003).
• How round is interpreted depends on the context
  in which it occurs round the block, round the
  barrier, round the corner, round the city center,
  
• Polysemy
• Prototype

11
Compositionality

• Context-dependency leads to problems of
  compositionality
• Principle of Compositionality of meaning the
  meaning of a complex whole is a function of the
  meaning of its parts and the way they are put
  together
• Underspecification?
• Prototype

12
Round in vector space semantics Zwarts and
Winter 2000, Zwarts 2003

• Completeness a vector pointing in every
  direction
• Constancy all the vectors are of the same length.

• Uniqueness do not touch a place more than once.

13
Violations of prototype

• Constancy not satisfied
• The earth goes around the sun.

• Uniqueness not satisfied
• The mouse keeps running round the computer

14
Weakening Completeness

• Inversion two vectors point in opposite
  directions (at least half-circle).
• The burglar drove round the barrier.

• Orthogonality two vectors point in perpendicular
  directions (at least a quarter-circle).
• He ran round the corner.

15
Strongest meaning wins

• Two contraints
• Strength favor stronger interpretations over
  weaker ones (Blutner 2000, Zeevat 2000)
• Fit consistency or avoid contradiction (Hendriks
  and de Hoop 2001, Zeevat 2000)
• Fit gtgt Strength
• Strong prototype has to be weakened to fit the
  context in which round occurs.

16
Possible interpretations

• Interpretations of round the door

17
round the door in OT
form round the door meaning Fit Strength
Completeness
? Inversion
Orthogonality
Detour
18
Context and Compositionality

• Local context and general context
• He walked round the door.
• He put his head round the door.
• Lexical semantics and world knowledge
• Bi-directional OT recoverability and
  compositionality.

19
Functional compositionality

• Blutner, Hendriks and de Hoop (2003) functional
  compositionality arises in a process of
  bi-directional optimization role of context is
  crucial.

20
Compositionality across languages

• Negation not, ?
• Negative quantifier nobody, nothing, ??
• Assumption knowledge of first-order logic part
  of general human cognition.
• Prediction negation, negative quantifiers behave
  alike across languages.
• Prediction is not borne out by data.

21
Quanta Costa?
22
Negative concord

• Nobody said nothing. (Eng) ??x??y
• Niemand zei niets. (Dutch) ??x??y
• Nadie miraba a nadie. (Spa) ??x?y
• Nessuno ha parlato con nessuno. (Ital)
  ??x?y
• Personne na rien dit. (Fr) ambiguous

23
Negation in context

• Double negation versus
• negative concord
• negative quantifiers versus n-words.
• In isolation same form, same meaning.
• In a sequence different forms, different
  meanings.
• Bi-directionality

24
Faithfulness and Markedness

• Negation is more marked than assertion, both in
  form and in meaning.
• Neg Avoid negation in the output.
• MaxNeg Mark the arguments of a negative chain
  formally.
• InterpretNeg Interpret every expression marked
  for negation as contributing a semantic negation.

25
Concord (production)

• MaxNeg gtgt Neg gtgt InterpretNeg
• Nessuno ha parlato con nessuno. (Ital)

meaning ??x?y form MaxNeg Neg IntNeg
negindef
? negneg
26
Concord (interpretation)

• MaxNeg gtgt Neg gtgt InterpretNeg

form negneg meaning MaxNeg Neg IntNeg
??x??y
? ??x?y
27
Double Negation (production)

• InterpretNeg gtgt Neg gtgt MaxNeg
• Niemand zei iets

meaning ??x?y form IntNeg Neg MaxNeg
? negindef
negneg
28
Double Negation (interpretation)

• InterpretNeg gtgt Neg gtgt MaxNeg
• Nobody said nothing

form negneg meaning IntNeg Neg MaxNeg
??x?y
? ??x??y
29
Various Rankings

• InterpretNeg gtgt Neg gtgt MaxNeg
• Multiple negative expressions ? multiple negation
  (English, Dutch, etc.)
• MaxNeg gtgt Neg gtgt InterpretNeg
• Multiple negative expressions ? single negation
  (Spanish, Italian, etc.)
• Overlapping range of Neg and InterpretNeg ?
  ambiguities (French).

30
Form and meaning

• If you mark arguments of a negative chain
  formally (MaxNeg gtgt Neg in production), make
  sure they are not interpreted as each
  contributing negation (Neg gtgt InterpretNeg in
  interpretation) (NC).
• If you interpret each negative expression as
  contributing a negation (InterpretNeg gtgt Neg in
  interpretation), make sure they are not marked
  formally as arguments of a negative chain (Neg
  gtgt MaxNeg in production) (DN).

31
Variation in form and meaning

• When meaning is less variable in context, we find
  variation in form.
• Ik begrijp er niks van/geen fluit/sikkepit/snars
• When form is less variable in context, we find
  variation in meaning.
• Hij eet een appel.
• He eats/is eating an apple.
• De Hoop, Haverkort en v.d. Noort (2003)

32
Moving on

• Bi-directionality and anaphora resolution elided
  meaning needs to be recoverable from context.
• Work on ellipsis and sluicing by Jennifer
  Spenader (Groningen)
• For an automatic system of pronoun resolution in
  OT, see G. Bouma (2003)

http//wodan.let.rug.nl/gerlof_bin/oplos_script
33
Language Acquisition

• Irene Krämer interpretation of quantifier many
  interpretation of pronouns and indefinites within
  a story context

34
Acquisition and Cognition

• Relation between learning of language and
  learning in other cognitive domains.
• Discourse children need to learn to use
  syntactic and discourse information over
  non-linguistic (contextual) information.

35
Conflicts in Interpretation

• OT modeling of interpretation
• Bi-directionality
• context-dependency
• Variation
• Language and learning
• Human cognition