Mapping between par;cles and inten;onality in Japanese preferen;al looking Franklin Chang, University of Liverpool Tessei Kobayashi, NTT Communica;on Sciences Laboratories, NTT Corp.
Overview Preferen;al looking paradigms have been used to examine the rela;onship between scenes and uferances (Naigles, 1990). But ogen scenes can vary in mul;ple ways (e.g., causa;vity, synchronicity, mo;on) and uferances can also differ in mul;ple ways (e.g., number of noun phrases, structure, role-‐func;on mapping). Here we seek to disentangle these features by using a language that allows us to match the number of noun phrases (Japanese) and scenes which are similar in mo;on and configura;on of par;cipants. Our study examines the rela;onship between Japanese par;cles and whether an ac;on is inten;onality caused.
Results Dependent measure: empirical logit of looking ;me to matching screen Mixed models with age (20, 24, 28), sec;on (baseline, linguis;c), and par;cle (SUBJ, OBJ) as predictors and subject and verb as random factors. Found significant three way interac;on (t=2.49, p < 0.02).
Proper&es of Japanese Japanese is a pro-‐drop language where phrases are marked by par;cles indica;ng their syntac;c func;on. apple OBJ hit -‐> “someone/he hit the apple” apple SUB fell -‐> “the apple fell” dog SUB hit -‐> “the dog hit something/it” Object par;cle is used to mark the object of an inten;onal ac;on.
20 months: No main effects or interac;ons. 24 months: Significant mismatch effect (t = 2.1, p < 0.04), where children look at the non-‐inten;onal video when they hear a phrase with an object par;cle. 28 months: Significant main effects of sec;on (t = 2.26, p <0.01), main effect of par;cle (t = 4.9, p < 0.001), and a significant interac;on (t = 4.6, p < 0.001). This is due to a significant baseline difference (t = 4.7, t < 0.001) due to the pairing of the novel verb and the matching screen in the exposure phase.
Our study Matched inten;onal ac;on and non-‐inten;onal ac;ons were used. Scenes differed in the par;cle that would normally be used. “duck-‐OBJ give” “duck-‐SUBJ fall”
Inten;onality was emphasized by showing two afempts for inten;onal ac;ons. A trial presented the scene twice and then a test phrase with a baseline looking phase and two presenta;ons of a linguis;c s;mulus. exposure phase pa “pateta” (novel verb)
Another mixed model by par;cle shows that the mapping between the SUBJ par;cle and meaning does not change over these ages. All development changes seem to be concentrated in the OBJ par;cle.
Findings These results did not conform to the predic;ons of theories of preferen;al looking. The significant mismatch effect for object par;cles is consistent with a number of noun phrases account, but we did not find a strong match effect for subject par;cles. The mismatch effect is difficult to explain in theories that assume that preferen;al looking indexes the match between scenes and innate/learned linguis;c knowledge.
Connec&onist Model of Preferen&al Looking Chang, Dell, & Bock (2006) proposed that error in predic;on from scene meaning could be used to explain preferen;al looking behavior. Error = Actual sequence – Predicted sequence from meaning Match Preference = Mismatch error/(Mismatch error + Match error) A Japanese version of the Dual-‐path model was created and trained on intransi;ves, transi;ves, and da;ves (like Chang, 2009). Model tested with transi;ve/intransi;ve message with a novel verb paired with single noun phrase with either subject or object par;cle.
baseline phase (4 secs) linguis;c phase (4 secs) “duck OBJ pateta” “duck OBJ pateta”
Object mismatch is greater than subject match
90 20-‐month, 24-‐month, and 28-‐month old Japanese children were tested.
Predic&ons If the number of noun phrases is cri;cal (Fisher, 2002), then since both phrases have only one noun phrase, there should be a strong match effect for subject par;cles and a mismatch effect for object par;cles. If the children have a link between the par;cles and inten;onality, then we would expect a match effect for both par;cles.
Model shows preference to match subject par;cle and transi;ve message, but there is a slight mismatch preference for object par;cles (below 0.5) early in development. Although there is a match preference at the par;cle, it is swamped by a mismatch preference at the noun, where two argument structures have more trouble predic;ng the first noun than one argument structures. This suggests that mismatch preference in this study could arise from conflict of cues at DIFFERENT sentence posi;ons as is suggested in Chang, et al. (2006).
References Chang, F. (2009). Learning to order words: A connec;onist model of heavy NP shig and accessibility effects in Japanese and English. Journal of Memory and Language, 61(3), 374–397. Chang, F., Dell, G. S., & Bock, J. K. (2006). Becoming syntac;c. Psychological Review, 113(2), 234–272. Fisher, C. (2002). Structural limits on verb mapping: the role of abstract structure in 2.5-‐year-‐olds' interpreta;ons of novel verbs. Developmental Science, 5(1), 55–64. Naigles, L. (1990). Children use syntax to learn verb meanings. Journal of Child Language, 17(02), 357–374.