280 likes | 429 Views
LSA 2008 Chicago, Illinois January 4, 2008. Elaine J. Francis Purdue University ejfranci@purdue.edu. The effects of weight on the processing of Extraposition from NP in English. Extraposition from NP (ENP) in English. In ENP, a noun is modified by a phrase that is apparently outside the NP:
E N D
LSA 2008Chicago, IllinoisJanuary 4, 2008 Elaine J. Francis Purdue University ejfranci@purdue.edu The effects of weight on the processing of Extraposition from NP in English
Extraposition from NP (ENP) in English • In ENP, a noun is modified by a phrase that is apparently outside the NP: • 1. Three people arrived here early yesterday morning who were from Chicago. (extraposed RC) • 2. Three people who were from Chicago arrived here early yesterday morning. (canonical RC) • Here are 3 of 358 real examples of ENP from the ICE-GB Corpus: • The best singer is this Olaf Bergh that I've seen. • We've got some friends coming to supper whose daughter's there so I can question tomorrow so I can question her about it. • I think it would take uh oh three-quarters of an hour to an hour for somebody to start who didn't have any any any experience at all.
ENP: A discontinuous dependency • Syntactic analyses include: • Rightward movement (Ross 1967, Baltin 1983) • Discontinuous NP (McCawley 1987, Hawkins 2004) • Base-generated adjunction to IP/VP (Rochemont & Culicover 1990) • Regardless of the analysis, it is clear that there is a discontinuous dependency between the NP and its extraposed modifier or complement, violating the normal phrase structure pattern of English NPs. • Why are ENP structures used and even preferred in some contexts over truth-conditionally equivalent canonical NP structures?
What licenses Extraposition from NP? • Syntactic factors(Baltin 1983; Guéron & May 1984; Rochemont & Culicover 1990; Culicover & Jackendoff 2005): • The grammar permits the word-order configuration in ENP and also appears to impose some constraints on its occurrence. • Discourse factors(Huck & Na 1992; Rochemont & Culicover 1990; Takami 1998): • Extraposed phrases tend to express discourse focus – new, important, or constrastive information. This is an instance of the more general principle Focus Last(Culicover & Jackendoff 2005). • Grammatical Weight(Wasow 2002; Hawkins 2004) • Extraposed phrases tend to be long/heavy relative to the preceding VP. This is an instance of the more general Principle of End Weight (Wasow 2002). • Principle of End Weight is motivated by processing factors in both perception and production.
Previous studies of weight effects • Extraposition from NP in German (Hawkins 2004; Uszkoreit et al 1998) • Heavy NP Shift in English (Stallings 1998; Wasow 2002) • Verb-particle constructions in English (Lohse, Hawkins, & Wasow 2004) • Topicalization in Cantonese (Matthews & Yeung 2001; Ching, Perry, Francis, & Matthews in prep) • No previous studies of weight effects in Extraposition from NP in English.
Research Questions • Is there a processing cost at all for the discontinuous dependency in ENP? • Can manipulating the weight of the postposed constituent improve processing efficiency and/or acceptability of ENP structures?
Processing Principle of Hawkins (2004) • Minimize Domains (MiD)-speakers and listeners prefer to minimize connected sequences of linguistic forms in which syntactic and semantic relations of combination and/or dependency are processed. • Phrasal Combination Domain (PCD)– the smallest string of elements required to construct a mother node (e.g., VP) and its immediate constituents.
PCDs for Canonical vs. ENP Structures Canonical RC within Subject NP as RC gets longer, matrix clause domain gets longer, NP domain stays the same [NP N [CP Relative Clause]][VP V (XP)] PCD for NP short PCD for MC long Extraposition from Subject NP as RC gets longer, matrix clause domain stays short, NP domain stays the same [NP N] [VP V (XP)] [CP Relative Clause] PCD for NP long PCD for MC short
Experiment 1: Self-paced Reading • Visual presentation of whole sentence on a computer screen (example of ENP structure with heavy RC): Two women showed up here at midnight who were in great distress about a missing dog that had run away from home. • Participants must press a button as soon as they have read and understood the sentence and then answer a comprehension question about the sentence. • Only accurate responses included in the analysis.
Weight-based Predictions • Reading times of ENP structures should get faster as RC weight increases. Advantage for ENP structure when RC is heavy (to minimize PCD for matrix clause). • Reading times of canonical structures should get slower as RC weight increases. Advantage for canonical structure when RC is light (to minimize PCD for Subject NP). • Reading times of adjunct clause should stay about the same (PCDs for Subject NP and matrix clause are not affected by weight of adjunct clause).
Design and Stimuli • 3x3 repeated measures design • Relative clause weight (words): light (4), medium (8), heavy (15) • Structure: canonical RC, extraposed RC, adjunct clause • 5 sets of 9 sentences (example of light condition): • Three peoplewho were from Chicagoarrived here early yesterday morning. (canonical RC) • Three peoplearrived here early yesterday morningwho were from Chicago. (extraposed RC) • Three peoplearrived here early yesterday morningafter they left Chicago. (adjunct clause) • 4 blocks of 33 sentences. 11 test sentences and 22 fillers in each block (12 test and 23 fillers in block 4). • Random ordering of sentences within each block, random ordering of blocks. • VP length held constant at 5 words.
Participants • 40 Purdue University students • Ages 18-39 (average age 22) • Native speakers of American English • 17 male, 23 female • Participants were each paid $6 for a 35-45 minute session.
Main effects: Weight: F(2, 38)= 6.56, p < 0.01 Interaction: Weight x Structure: F(4, 36)= 3.18, p = 0.02Planned t-tests: Heavy Canonical vs Heavy Extraposed: t = 2.59, p = 0.01 Light Extraposed vs. Heavy Extraposed: t = 4.77, p < 0.01 Light Canonical vs. Heavy Canonical: t = 1.33, p = 0.19 (n.s.)
Experiment 1 Summary • Advantage for ENP structure over canonical structure in heavy condition: • For ENP structures, mean reading time per word got faster as clause weight increased, as predicted. • In the heavy condition, ENP structures were read significantly faster than canonical structures. • In the light and medium conditions, there was no significant difference between ENP and canonical structures, though canonical structures were read slightly faster. • As predicted, RTs for adjunct clause did not change as a function of weight. • Contrary to prediction, RTs for canonical structure did not get slower as a function of weight.
Experiment 2: Acceptability Judgment Task • Written survey asking for acceptability ratings for each sentence on 9-point scale: Two women showed up here at midnight who were in great distress about a missing dog that had run away from home. WorstBest 1 2 3 4 5 6 7 8 9
Weight-based Predictions • Ratings of ENP structures should get better as RC weight increases. Advantage for ENP structure when RC is heavy (to minimize PCD for matrix clause). • Ratings of canonical structures should get worse as RC weight increases. Advantage for canonical structure when RC is light (to minimize PCD for Subject NP). • Ratings of adjunct clause should stay about the same (PCDs for Subject NP and matrix clause are not affected by weight of adjunct clause).
Design and Stimuli • 3x3 repeated measures design • Relative clause weight (words): light (4), medium (8), heavy (15) • Structure: canonical RC, extraposed RC, adjunct clause • 5 sets of 9 sentences (example of light condition): • Three peoplewho were from Chicagoarrived here early yesterday morning. (canonical RC) • Three peoplearrived here early yesterday morningwho were from Chicago. (extraposed RC) • Three peoplearrived here early yesterday morningafter they left Chicago. (adjunct clause) • 4 blocks of 33 sentences. 11 test sentences and 22 fillers in each block (12 test and 23 fillers in block 4). • Four survey scripts with two different orderings of sentences within each block, two different orderings of blocks. Pseudo-random order to avoid similar sentences on the same page. • VP length held constant at 5 words.
Participants • 10 Purdue University students • Ages 20-54 (average age 24) • Native speakers of American English • 5 male, 5 female • Participants were paid $6 for a 35-45 minute session. • Experiment still in progress. We plan to recruit 20 more participants.
Main effects: Weight: F(2, 8)= 1.5, p = 0.28 (n.s.) Structure: F(2, 8) = 2.61, p = 0.13 (n.s.)Interaction: Weight x Structure: F(4, 6)= 3.54, p = 0.08 (marginal)Planned t-tests: Light Canonical vs Light Extraposed: t = 2.77, p = 0.02 Heavy Canonical vs Heavy Extraposed: t = 0.11, p = 0.91 (n.s.) Light Extraposed vs. Heavy Extraposed: t = 0.23, p = 0.82 (n.s.) Light Canonical vs. Heavy Canonical: t = 3.59, p < 0.01
Main effects: Weight: F(2, 8)= 0.43, p = 0.66 (n.s.) Structure: F(2, 8) = 3.03, p = 0.10 (n.s.)Interaction: Weight x Structure: F(4, 6)= 15.8, p < 0.01Planned t-tests: Light Canonical vs Light Extraposed: t = 2.54, p = 0.03Heavy Canonical vs Heavy Extraposed: t = 0.00, p = 1.0 (n.s.) Light Extraposed vs. Heavy Extraposed: t = 1.15, p = 0.28 (n.s.) Light Canonical vs. Heavy Canonical: t = 3.28, p < 0.01
Experiment 2 summary • Advantage for canonical structure over ENP structure in light and medium conditions, which disappears in heavy condition. • For canonical structures, ratings decreased as clause weight increased, as predicted. However, this decrease only happened in the heavy condition. • In the light and medium conditions, canonical structures were rated significantly higher than ENP structures. • In the heavy condition, there was no significant difference between ENP and canonical structures, unlike in Expt. 1. • As predicted, ratings for adjunct clauses did not change as a function of weight. • Contrary to prediction and unlike in Experiment 1, ratings for ENP structure did not significantly increase as a function of weight.
Conclusions • Reading Time • Processing efficiency of ENP structures increased as a function of grammatical weight, as shown by faster reading times. • Sentences with heavy RCs were read faster when RC was extraposed, whereas sentences with light RCs were read at about the same speed in both structures. • No additional processing cost for ENP compared to canonical structure. • Acceptability • Acceptability of canonical structures decreased as a function of weight. • Sentences with light RCs were judged more acceptable with canonical structure, whereas sentences with heavy RCs were judged equally acceptable in both structures. • Grammatical weight is a significant factor in both comprehension and acceptability of ENP structures in English.
Future Research • Comprehension vs. Acceptability • Why did weight affect reading time differently than acceptability judgments? • Corpus study • Are ENP structures used more frequently when the extraposed phrase is heavy? • Production study • Does grammatical weight affect planning and production of ENP structures? • Grammatical Weight vs. Discourse Focus • Does discourse information structure contribute to processing, usage patterns, and/or acceptability of ENP independently of grammatical weight?
Thanks to • Research assistants: Najeong Kim and Yanhong Zhang • Funding: Purdue University, College of Liberal Arts & Department of English • Experiment participants at Purdue • Colleagues who provided comments • LSA abstract reviewers, conference organizers, and audience
Selected References • Baltin, Mark R. 1983. Extraposition: bounding versus government-binding. Linguistic Inquiry`14(1): 155-162. • Ching, Hoi Lam, Conrad Perry, Elaine J. Francis, and Stephen Matthews. in prep. The advantages of topicalization: An experimental study of sentence processing in Cantonese. • Cowart, Wayne. 1997. Experimental Syntax. Thousand Oaks, CA: Sage Publications. • Culicover Peter.W. and Ray S. Jackendoff. 2005. Simpler Syntax. Oxford: Oxford University Press. • Guéron, Jacqueline and Robert May. 1984. Extraposition and logical form. Linguistic Inquiry 15(1): 1-30. • Hawkins, John A. 2001. Why are categories adjacent? Journal of Linguistics 37: 1-34. • Hawkins, John A. 2004. Efficiency and Complexity in Grammars. Oxford: Oxford University Press. • Huck, Geoffrey J. and Youngnee Na. 1992. Information and contrast. Studies in Language 16(2): 325-334. • Lohse, Barbara L., John A. Hawkins, and Thomas Wasow. 2004. Domain minimization in English verb-particle constructions. Language 80(2): 238-261.
References, continued • Matthews, Stephen and Yeung, Louisa Y.Y. 2001. Processing motivations for topicalization in Cantonese. In Kaoru Horie & Shigeru Sato, eds. Cognitive-functional Linguistics in an East Asian Context, 81-102. Toyko: Kurosio. • McCawley, James D. 1987. Some additional evidence for discontinuity. In Geoffrey J. Huck and Almerindo E. Ojeda, eds. Discontinuous Constituency, 185-200. Orlando, FL: Academic Press. • Rochemont, Micahel S. and Peter W. Culicover. 1990. English Focus Constructions and the Theory of Grammar. Cambridge: Cambridge University Press. • Takami, Ken-ichi. 1998. A functional constraint on Extraposition from NP. In Akio Kamio and Ken-ichi Takami, eds. Function and Structure, 23-56. Amsterdam: John Benjamins. • Uszkoreit, Hans, Thorsten Brants, Denys Duchier, Brigitte Krenn, Lars Konieczny, Stephan Oepen, and Wojciech Skut. 1998. Studien zur performanzorientierten Linguistik: Aspekte der Relativsatzextraposition im Deutschen. Kognitionswissenschaft 7(3): 129-133. • Wasow, Thomas. 1997. End-weight from the speaker's perspective. Journal of Psycholinguistic Research 26: 347-361. • Wasow, Thomas. 2002. Postverbal Behavior. Stanford: CSLI Publications.