Incremental Processing of Gap-filler Dependencies: Evidence from the Processing of Subject and Object Clefts in Japanese Barış KAHRAMAN (1), Atsushi SATO (1), (2), Hajime ONO (3) , & Hiromu SAKAI (1) (1)Hiroshima University; (2)Japan Society for the Promotion of Science; (3) Kinki University Correspondence: [email protected] (Barış Kahraman)

INTRODUCTION BACKGROUND

PREVIOUS STUDIES  In head-final languages the processing asymmetry of SR and OR was observed at different positions: SUPPORTS Korean: RC-verb & filler (Kwon, 2008) INCREMENTALITY Turkish: RC-verb (Kahraman et al, 2010)

FILLER-GAP DEPENDENCIES The teacher who i [the student saw GAP

FILLER

SEARCH

i]

was young.

GAP

 Filler-gap dependencies are processed incrementally: When the filler precedes the gap, the parser posits a gap as soon as the filler is encountered and starts a search for that gap (e.g. Aoshima et al, 2004; Stowe 1986)

GAP-FILLER DEPENDENCIES  A dependency in which the gap precedes the filler such as relative clauses (RCs) in head final languages (Kwon, 2008). [Gakusei-ga senseii –wa wakakatta. i mita] [Student-NOM

i

GAP

saw]

 Japanese RCs are structurally ambiguous between a matrix clause and a subordinate clause at RC-verb. ⇒  Since the Japanese parser cannot determine whether a filler will appear, it might have not established a gap-filler dependency until the filler appears.  What if the parser predicts the upcoming filler earlier?

PRESENT STUDY  Japanese clefts provide a good test case to examine the incremenatlity of gap-filler dependency formation.

teacheri –TOP young.

[Gakusei-ga/o

FILLER SEARCH

DOES NOT SUPPORT INCREMENTALITY

Japanese: Filler (e.g. Ishizuka, 2005)

?

FILLER

 Are the gap-filler dependencies also processed incrementally? Does the parser posit a filler when it identifies the gap, and start to construct a dependency prior to the filler’s appearance?

mita-no-wa]

suugaku-no senseii –da.

[Student-NOM/ACC saw-COMP-TOP] math-GEN teacheri –COP `It was the math teacher who / the student saw / saw the student.`

 Unlike RCs, no-wa (cleft-marker) is attached to the verb. ⇒Structural ambiguity is resolved at the embedded verb and existence of an upcoming filler is signaled. PREDICTION: If the Japanese parser immediately posits a filler and  It is assumed that the processing asymmetry between subject and starts a search for it from the verb position, the processing asymmetry object relative clauses (SR / OR) reflects the relative ease / difficulty between subject and object clefts (SC / OC) would be observed at the of establishing a filler-gap dependency (e.g. Gibson, 1998; O’Grady, verb (Kahraman et al, 2010; Kwon, 2008). 1997).  AIM: In order to test our prediction, we conducted a self-paced  It is conceivable that the observed position of the processing reading experiment, and examined whether the parser starts to asymmetry indicates that the formation process of a gap-filler construct a gap-filler dependency before the filler appears. dependency has taken place at that position.

EXPERIMENTS EXPERIMENT 1

EXPERIMENT 2

Aim: Using SRs and ORs, to examine the reliability of test items. Participants: 36 native speakers of Japanese at Hiroshima University. Materials: 30 set of SRs & ORs + 66 fillers

Aim: To test whether there is a processing asymmetry between SCs and OCs. If so, where is the processing asymmetry observed? Participants: 26 native speakers of Japanese at Hiroshima University. Materials: 30 set of SCs & OCs + 80 fillers

SR: 1 2 3 4 5 6 7 8 Kyonen sobo-o inaka-de kaihooshita tooi shinseki-wa jiko-de nakunatta Last year grandma-ACC village-LOC nursed distant relative accident-in died ‘The distant relative who nursed the grandma last year at village died in an accident.’

SC: 1 2 3 4 5 6 7 8 Kyonen sobo-o inaka-de kaihooshita-nowa tooi shinseki-da-to haha-ga itta. Last year grandma-ACC village-LOC nursed-NOWA distant relative-COP-that mom-NOM said My mother said that it was the distant relative who nursed the grandma last year at village.

OC: 1 2 3 4 5 6 7 8 OR: 1 2 3 4 5 6 7 8 Kyonen sobo-ga inaka-de kaihooshita tooi shinseki-wa jiko-de nakunatta Kyonen sobo-ga inaka-de kaihooshita-nowa tooi shinseki-da-to haha-ga itta. Last year grandma-NOM village-LOC nursed distant relative accident-in died ‘The distant relative who the grandma nursed last year at village died in an accident.’

Predictions: If there is no problem with items, SRs should be read faster than ORs at the head-noun (e.g. Ishizuka, 2005; Miyamoto & Nakamura, 2003; Ueno & Garnsey, 2008).

Last year grandma-NOM village-LOC nursed-NOWA distant relative-COP-that mom-NOM said My mother said that it was the distant relative who the grandma nursed last year at village.

Predictions: If the parser starts to construct a gap-filler dependency before the filler, (1) processing asymmetry would be observed at the Verb, and (2) SCs would be read faster than OCs (Gordon et al, 2001).

RESULTS R T (m s ) 750

RESULTS

[F1 (1,35) = 4.85, p < .05; F2 (1,29) = 3.84, p = .06]

9 0 0

650

*

600 500 450 r e g io n 1

[F1 (1,25) = 9.76, p < .01; F2 (1,29) = 5.12, p < .05]

1 0 0 0

700

550

R T ( m s) 1 1 0 0

8 0 0 7 0 0

SR OR 2

6 0 0

O C :

5 0 0 3

4 RC -v erb

5

6 f ille r

7

8

 The head-noun of SRs was read faster than that of ORs.  We replicated the previous studies in Japanese (e.g. Ishizuka, 2005)  Our test items are reliable.

r e g io n

**

S C :

1

2

3

4 ve rb

5

6 f ille r

7

8

 Participants read OCs faster than SCs.  The processing asymmetry was observed at the embedded verb.

GENERAL DISCUSSION & CONCLUSIONS  Processing ease of RCs and clefts cannot be explained by the same structural factors (i.e. Structural distance)  RCs and clefts were processed differently in Japanese.  Why does the processing ease of RCs and clefts differ in Japanese? 1) PROCESSING ASYMMETRY POSITION: Although the processing POSSIBLE FACTORS: ease of RCs and clefts were different, the processing asymmetry 1) Discourse function of RCs (topic) and clefts (focus) might be related reflects the formation ease of gap-filler dependencies. to their processing ease (Roland, 2009). RCs: At the head-noun (filler) ⇒ 2) Frequency of SRs vs. ORs & SCs vs. OCs might be different (Reali & Christiansen, 2007)  Gap-filler dependency formation was established at the filler 3) Prediction for the upcoming structures at the embedded verb might Clefts: At the embedded verb ⇒ be different between RCs and Clefts (Hale, 2006; Levy, 2008).  Gap-filler dependency formation was established at the verb.  In the future studies, we will attempt to explain possible factors.  The cleft marker has an impact on the processing of gap-filler dependency CONCLUSIONS  In clefts, the gap-filler dependency is processed incrementally.  In addition to the filler-gap dependencies, the gap-filler 2) PROCESSING EASE of RCs & CLEFTS: dependencies are also processed incrementally (e.g. Aoshima et al., RCs ⇒ SRs < ORs 2004; Stowe 1986).  Consistent with previous studies in Japanese and other languages  The use of cleft-marker would be an important source for the Clefts ⇒ OCs < SCs incremental processing of gap-filler dependencies (Kahraman et al.,  Inconsistent with previous studies in English (e.g. Gordon et al., 2001). 2010)

GENERAL DISCUSSION

TCPTCP-2011 March 1111-12 KEIO UNIVERSITY

ACKNOWLEDGEMENT: This research was supported by (1)Grant-in-Aid for Scientific Research (B) “Neurocognitive basis for language learning through the processing of input and output (PI: Hiromu Sakai, #20320060) ” by JSPS; (2) Grant-in-Aid for Scientific Research (B) (PI: Kentaro Nakatani, #21320083) by JSPS; (3) Grant-in Aid for Young Scientists from the MEXT (PI: Hajime Ono, # 21720152).

Incremental Processing of Gap-filler Dependencies

between subject and object clefts (SC / OC) would be observed at the ... R T ( m s ). S C : O C : **. [F1 (1,25) = 9.76, p < .01; F2 (1,29) = 5.12, p < .05].

183KB Sizes 2 Downloads 72 Views

Recommend Documents

Approximate and Incremental Processing of Complex ...
propose a new approach, allowing an affordable computation of an ini- tial set of ... The success of current Web search engines suggest that exact matching ...... formed without optimization, i.e., based on fixed query plans (same for both.

Large-scale Incremental Processing Using ... - Research at Google
language (currently C++) and mix calls to the Percola- tor API with .... 23 return true;. 24. } 25. } 26 // Prewrite tries to lock cell w, returning false in case of conflict. 27 ..... set of the servers in a Google data center. .... per hour. At thi

Factors in the incremental processing of NPs in Japanese
The bulk of the evidence has come from Japanese. The problem with verb ..... they able to infer that 'Ken' acted on an 'email' (e.g., read/received/sent/wrote)?.

Complimentary aspects of Incremental & Radical ...
In contrast, established firms devote a small fraction of their resources to trying ... next-generation business model, create a buzz in the boardroom while lesser forms of ... computer workstations; the Java operating system, which has driven much i

Incremental Learning of Nonparametric Bayesian ...
Jan 31, 2009 - Conference on Computer Vision and Pattern Recognition. 2008. Ryan Gomes (CalTech) ... 1. Hard cluster data. 2. Find the best cluster to split.

Outage Performance of Multi-Antenna Cooperative Incremental ...
Email: [email protected] ... the link from source to destination while the other locates near ... as an incremental link, and this system is referred to as.

Complimentary aspects of Incremental & Radical ...
Established companies spend 80-90 percent of their technology budgets on ... software and consulting, which rivals have used to lock in customers. ... In cricket the best strategy is to have one batsman take risk and go after not so loose.

Generalized File System Dependencies - SOSP 2007
Oct 14, 2007 - A patch represents both a change to disk data and any dependen- ... Permission to make digital or hard copies of all or part of this work for.

Secure Dependencies with Dynamic Level ... - Semantic Scholar
evolve due to declassi cation and subject current level ... object classi cation and the subject current level. We ...... in Computer Science, Amsterdam, The Nether-.

Framing Dependencies Introduced by Underground ... - Damon McCoy
International Computer Science Institute .... and ever increasing degrees of specialization. ..... variants that can be cleaned up by purchasing a one-year sub-.

On linguistic representation of quantitative dependencies
bInstitute of Problems of Informatics, Academy of Sciences of Tatarstan, .... year to year but the seasonal changes have usually the ... INCREASING in rule (1) may be interpreted as a linguistic ..... degree of porosity loss', 'Dominantly chemical pr

Framing Dependencies Introduced by Underground ... - Damon McCoy
and fragile homogeneous software systems provided fertile ground for the .... which primarily exist to support traditional crime using online distribution channels. [21] ... the spammer first requires a multitude of fake accounts to post messages.