Another word on parsing relative clauses Eyetracking evidence from Spanish and English
description
Transcript of Another word on parsing relative clauses Eyetracking evidence from Spanish and English
Another word on parsing relative clausesEyetracking evidence from Spanish and English
Manuel Carreiras & Charles Clifton, Jr.
Universal Parsing Strategies?
Preference for the simplest interpretation Minimal attachment strategy (Frazier, 1979, 1987)
• Not postulating any unnecessary nodes Late closure
• Attach new items into clause/phrase currently processing
Are these strategies universal? Cuetos & Mitchell (1988) – late closure fails to apply
in Spanish to parsing of RCs preceded by complex NPs
1. Someone shot the female servant of the actress who was on the balcony
Universal Parsing Strategies?
Preference for the simplest interpretation Minimal attachment strategy (Frazier, 1979, 1987)
• Not postulating any unnecessary nodes Late closure
• Attach new items into clause/phrase currently processing
Are these strategies universal? Cuetos & Mitchell (1988) – late closure fails to apply
in Spanish to parsing of RCs preceded by complex NPs1. Someone shot the female servant of the actress who
was on the balcony
Universal Parsing Strategies?
Preference for the simplest interpretation Minimal attachment strategy (Frazier, 1979, 1987)
• Not postulating any unnecessary nodes Late closure
• Attach new items into clause/phrase currently processing
Are these strategies universal? Cuetos & Mitchell (1988) – late closure fails to apply
in Spanish to parsing of RCs preceded by complex NPs1. Someone shot the female servant of the actress who
was on the balcony
Universal Parsing Strategies?
Preference for the simplest interpretation Minimal attachment strategy (Frazier, 1979, 1987)
• Not postulating any unnecessary nodes Late closure
• Attach new items into clause/phrase currently processing
Are these strategies universal? Cuetos & Mitchell (1988) – late closure fails to apply in
Spanish to parsing of RCs preceded by complex NPs1. Someone shot the female servant of the actress who was
on the balcony2. Alguien disparó contra la criada de la actriz que estaba en
el balcón
Universal Parsing Strategies?
Preference for the simplest interpretation Minimal attachment strategy (Frazier, 1979, 1987)
• Not postulating any unnecessary nodes Late closure
• Attach new items into clause/phrase currently processing
Are these strategies universal? Cuetos & Mitchell (1988) – late closure fails to apply in
Spanish to parsing of RCs preceded by complex NPs1. Someone shot the female servant of the actress who was
on the balcony2. Alguien disparó contra la criada de la actriz que estaba en
el balcón
Universal Parsing Strategies?
Preference for the simplest interpretation Minimal attachment strategy (Frazier, 1979, 1987)
• Not postulating any unnecessary nodes Late closure
• Attach new items into clause/phrase currently processing
Are these strategies universal? Cuetos & Mitchell (1988) – late closure fails to apply in
Spanish to parsing of RCs preceded by complex NPs1. Someone shot the female servant of the actress who was
on the balcony2. Alguien disparó contra la criada de la actriz que estaba en
el balcón
Universal Parsing Strategies?
Differences across other languages? In Spanish, French, German, and Dutch, the head of
the complex NP (N1) is preferred as the subject of the RC
• See Carreiras & Clifton (1999), p. 827 for complete list Italian readers initially prefer N2 as the agent
• DeVincenzi & Job, 1993, 1995 As other studies have shown, English readers either
prefer N2 as the agent of the RC or show no preference
• Carreiras & Clifton, 1993; Henstra, 1996
Universal Parsing Strategies?
An experimental artifact? Gilboy & Sopena (1996) - Segmentation
• Obtained preference for high attachment of N1 to RC only with large segmentation
• La policía arrestó a la hermana del criado/que dio a luz recientemente a dos gemelos
• The police arrested the sister of the handyman/who recently gave birth to twins
• No effects were found for small segmentation (splitting RC into two displays)
• La policía arrestó/a la hermana/del criado/que dio a luz recientemente a dos gemelos
• The police arrested/the sister/of the handyman/who recently gave birth to twins
• Conclusion: N1 preferences only arise when a particular segmentation (large) allows for characteristics of prosodic patterns to appear
Purpose
Two-fold:1. Whether N1 preference in Spanish is a byproduct of
segmentation or an underlying property of the language
2. Whether native English and native Spanish readers resolve ambiguity of attachment of RCs preceded by NPs in different ways for the same sentences
Examined performances for ambiguous structures of the type: N1 of N2 RC
Three eye-movement studies:1. Experiments 1&2: conducted in Spanish2. Experiment 3: conducted in English
Experiment 1
Subjects 44 undergraduate students
Apparatus Sentences were presented in lowercase letters on a
monitor which displayed up to 80 characters per line
Eye movements were monitored by a Dual Purkinje Eyetracker
Experiment 1
Design 16 sentences (English/Spanish) which contained a
complex NP (N1 de N2) followed by an RC intermixed with 144 other filler sentences
RC attachment was disambiguated by gender information
Experiment 1
Results (First-Pass Times) No significant effects for:
1. First-pass times at CR• Those disambiguated toward high attachment were
numerically faster
2. Masculine and feminine hosts3. Interaction between type of host and type of
disambiguation
Experiment 1
Results (Total time) Ss read CRs more rapidly when disambiguating
toward high attachment (N1) Sentences requiring feminine hosts for
disambiguation were read faster than those with masculine hosts
Conclusion: high attachment preferences are REAL
Experiment 2
Purpose High attachment was numerically present only when
disambiguating part of the RC required a masculine antecedent
• May be that more masculine than feminine RCs contained disambiguating morphology
Wanted to examine different sentences, all of which were disambiguated morphologically by gender
Experiment 2
Results (First-pass) No significant effects for:
• Type of disambiguation closure• Type of host• Interaction between disambiguation and host
Results (Total time) Regions disambiguating toward high attachment (N1)
were read faster than those toward low attachment (N2) Conclusion: effect not restricted to a particular gender
Experiment 3
Purpose Determine whether English readers have a bias to
interpret an RC as modifying the most recent noun (N2)
Method Subjects: 36 undergraduate students Same apparatus and design as Experiment 1
Experiment 3
Results (First-Pass) Ss read the CR faster when disambiguated toward
low attachment (N2) No significant effects for type of host or interaction
Results (Total Time) Low attachment disambiguations (N2) were read
faster than high attachment ones (N1)
Conclusions
Spanish readers have a modest preference for interpreting an RC as modifying NP1 and is not a consequence of segmentation
American English readers show a preference for low attachment of an RC to N2, although this result has not always been found
Implications
Why do the 2 languages differ?1. Spanish does not violate closure (Cuetos & Mitchell,
1988) Low attachment is an expression of a universal
processing principle, such as late closure, which is overridden by certain constructions of other languages
2. Tuning hypothesis (Mitchell & Cuetos, 1991) Initial resolution of structural ambiguities is
determined by the frequency with which alternative disambiguations are encountered
3. Predicate-proximity/Recency (Gibson, et al., 1996) Initial preferences are guided by weights of
parameters, that may differ among languages: Predicate proximity (attach to head of predicate) Recency (attach to most recent site)
Implications
Why do the 2 languages differ?1. Existence of the Saxon genitive (‘s)
Limits late closure to primary phrases (e.g. the colonel’s daughter who was on the porch)
If a speaker wanted to express high attachment, he or she would use the ‘s form
2. Pronoun interpretation (Hemforth, Konieczny, & Scheepers) RCs are associate in languages such as German by
default with the most salient available host (generally N1, the head of the complex NP and argument of main predicate)
In English, the word that is often used to introduce RCs, causing RCs to be treated less like pronouns and more like complement phrases (obeying late closure)
Implications
Why do the 2 languages differ?1. Heavy RCs
RCs are generally heavy, and prefer a relatively large host phrase
High attachment makes the N1 of N2 phrase the host (closer in size to typical RC than just N2 alone)
In English, because the word that is often phonologically reduced, it is absorbed into the preceding prosodic phrase
Thus, N2 is relatively heavier and the RC is correspondingly less heavy, encouraging their attachment