Corpus Frequency of Relative Clause Association in Japanese [isi-ga - PDF document

Relative Clause Association Ambiguity (1) [ 医師が触診している ] 少女の兄 (global structural ambiguity) Corpus Frequency of Relative Clause Association in Japanese [isi-ga syokusinsiteiru] syôzyo-no ani doctor-NOM palpating girl-GEN brother Toshiyuki YAMADA [RC] N1 N2 Douglas ROLAND (two association sites available after the RC) Manabu ARAI ‘the brother of the girl [(that) the doctor is palpating]’ Yuki HIROSE NP1 NP2 [RC] (The University of Tokyo) ( attachment ambiguity : two attachment sites available prior to the RC) TL/MAPLL2014 @ UT 1/32 TL/MAPLL2014 @ UT 0. Overview 2/32 August 13th, 2014 August 13th, 2014 Puzzle: Unforced Revision? Possibility: Structural Frequency Bias � If we believe in Revision as Last Resort (Fodor & Frazier, 1980) , � RC association ambiguity: the N1 association should be maintained. [RC] N1-GEN N2 N2 Association N1 Association N1 Association � RC S N1 N2 Does a structural frequency bias lead to the N2 association preference? [isi-ga syokusinsiteiru] syôzyo-no ani doctor palpating girl-GEN brother � Unforced revision to N2 association (Yamada et al., 2014) � Why is the N2 association preferred? TL/MAPLL2014 @ UT 0. Overview 3/32 TL/MAPLL2014 @ UT 0. Overview 4/32 August 13th, 2014 August 13th, 2014 Roadmap Structural Frequency Effects 0. Overview � RC attachment ambiguity (Cuetos & Mitchell, 1988) 1. Introduction NP1 of-NP2 [RC] 2. Yamada et al. (2014) - native English speakers: low attachment preference - native Spanish speakers: high attachment preference 3. The current study 4. Discussion � Corpus frequency (Mitchell et al., 1992) 5. Conclusions - English: 62% for low attachment � low preferred - Spanish: 60% for high attachment � high preferred � Frequency effects in PP attachment (e.g., Katsika, 2011 for Greek, cf. Gibson et al., 1996 for English) TL/MAPLL2014 @ UT 5/32 TL/MAPLL2014 @ UT 1. Introduction 6/32 August 13th, 2014 August 13th, 2014

Previous Findings on RC Association On-line N2 Association Preference (2a) N2-incompatible [Isi-ga tatta ima syokusinsiteiru] syôzyo-no ani-ga baiten-de kaimonositeiru. � Mixed results doctor-NOM just now palpating girl-GEN brother-NOM store-at shopping - N1 association preference (on-line evidence) ‘The brother of the girl that the doctor is palpating just now is shopping at a store.’ (2b) N2-incompatible - N2 association preference (off-line evidence) [Isi-ga tatta ima syokusinsiteiru] kanzya-no ani-ga baiten-de kaimonositeiru. (e.g., Kamide & Mitchell, 1997; Miyamoto et al., 2004; Uetsuki, 2007; Nakano, 2008) doctor-NOM just now palpating patient-GEN brother-NOM store-at shopping ‘The brother of the patient that the doctor is palpating just now is shopping at a store.’ � a possibility: on-line revision (2c) N2-compatible N1 association � N2 association [Isi-ga tatta ima syokusinsiteiru] syôzyo-no ani-ga isu-de zittositeiru. doctor-NOM just now palpating girl-GEN brother-NOM chair-at sitting still � Yamada et al. (2014): suggestive eye-tracking evidence ‘The brother of the girl that the doctor is palpating just now is sitting still on a chair.’ (cf. Bai et al., 2014: no commitment to the N1 association) (2d) N2-compatible [Isi-ga tatta ima syokusinsiteiru] kanzya-no ani-ga isu-de zittositeiru. doctor-NOM just now palpating patient-GEN brother-NOM chair-at sitting still ‘The brother of the patient that the doctor is palpating just now is sitting still on a chair.’ TL/MAPLL2014 @ UT 1. Introduction 7/32 TL/MAPLL2014 @ UT 2. Yamada et al. (2014) 8/32 August 13th, 2014 August 13th, 2014 Unforced Revision Research Question 1 � [RC] N1-GEN N2 Is there a structural frequency bias towards - initial commitment to the N1 association the N2 association? - revision to the N2 association � Prediction � Although the N1 association is grammatical, Japanese comprehenders are willing to make an a strong frequency bias towards the N2 association unforced revision to the N2 association at the point of when N1 and N2 associations are both plausible, the N2. situation most similar to the experimental items in � processing difficulty when the sentence-final main Yamada et al. (2014) predicate is incompatible with the N2 association TL/MAPLL2014 @ UT 2. Yamada et al. (2014) 9/32 TL/MAPLL2014 @ UT 3. The Current Study 10/32 August 13th, 2014 August 13th, 2014 Method Data Analysis: Ambiguous RC Association � 押収した山林の土 � - corpus: Kyoto University Text Corpus (5) Both N1 and N2 associations are plausible (http://nlp.ist.i.kyoto-u.ac.jp/EN/index.php?Kyoto%20University%20Text%20) � 38,400 samples (i.e., sentences) in total � oosyuusita sanrin-no tuti � seized forest-GEN soil - target samples: verb + N1 + no (GEN) + N2 ‘� the soil of the forest that was seized �’ � 4034 samples collected with restrictions, e.g., 1. verb dependent on either N1 or N2 2. nouns could not be tame ‘for’, koto ‘the fact (that �)’, nado ‘etc.’, naka ‘inside’ - procedure: false positive matches were excluded � 861 samples remained for analysis TL/MAPLL2014 @ UT 3. The Current Study 11/32 TL/MAPLL2014 @ UT 3. The Current Study 12/32 August 13th, 2014 August 13th, 2014

Data Analysis: Forced RC Association Data Analysis: N1 or N2 Association この外交の意味と原則を理解しない政治家の言動 � 「オンリーワン」哲学を実践してきた三井氏の答え � (6) a. N1 association forced (7) a. N1 association intended Kono gaikou-no imi to gensoku-o rikaisinai seizika-no gendou � “Onrii wan” tetugaku-o zissensitekita Mitui-si-no kotae � this diplomacy-GEN meaning and principle-ACC understand.not politician-GEN behavior only one philosophy-ACC have.practiced M.-Mr.-GEN answer ‘The behavior of the politician that does not understand the ‘The answer of Mr. Mitsui that has practiced “Only one” 時代の変化に合った事業の見直し � meaning and principle of this diplomacy �’ philosophy �’ 低く抑えた男の声 � b. N2 association intended b. N2 association forced Zidai-no henka-ni atta zigyou-no minaosi � Hikuku osaeta otoko-no koe � times-GEN change-for suitable project-GEN revision ‘the revision of the project that is suitable for the change low made male-GEN voice ‘The voice of the male that is made low �’ accompanying the times �’ TL/MAPLL2014 @ UT 3. The Current Study 13/32 TL/MAPLL2014 @ UT 3. The Current Study 14/32 August 13th, 2014 August 13th, 2014 Frequency did not seem to explain RC But, there was a N2 bias when only association preferences in Japanese ambiguous examples were considered Frequency bias: Comprehenders may take ambiguity into � English account in generating � Spanish RC N1 の RC N1 の association � Japanese preferences. N2 N2 (GEN) (GEN) Association Association N1 N2 N1 N2 All examples 408 427 Ambiguous examples only 237 336 TL/MAPLL2014 @ UT 3. The Current Study 15/32 TL/MAPLL2014 @ UT 3. The Current Study 16/32 August 13th, 2014 August 13th, 2014 Weak Bias towards the N2 Association Research Question 2 � Finding 1 Do we observe the N2 bias when the situation is no bias overall, but most similar to the experimental items in Yamada N2 bias when both associations were plausible et al. (2014)? � Contrary to the prediction, � Several interacting factors might affect comprehenders’ the bias towards the N2 association was not sufficiently processing of [RC] N1-GEN N2 strong to override the semantic preference for N1 - length of RC (short vs. long) (e.g., Hirose et al., 1998) association in Yamada et al. (2014) - RC-type (subject- vs. object-extracted) (e.g., Uetsuki, 2006) - animacy of N1 and N2 � The structural frequency bias is not the only source, but - properties of N1 and N2 comprehenders may take ambiguity into account in generating association preferences (i) discourse status and (ii) uniqueness TL/MAPLL2014 @ UT 4. Discussion 17/32 TL/MAPLL2014 @ UT 3. The Current Study 18/32 August 13th, 2014 August 13th, 2014