The predictability effect on eye movement in reading Korean dative sentences

( Hongoak Yun ),( Dongsu Lee ),( Yunju Nam ),( Upyong Hong ) 한국언어정보학회 2017 언어와 정보 Vol.21 No.1

Yun and Hong (2014) demonstrated the effect of role predictability and word predictability in the processing of Korean dative sentences. We aimed to investigate the nature of predictability effect that Yun and Hong observed, in particular, to explore whether the effect would reflect the early stage or the late stage of processing by observing readers` eye movement in reading. The significant effect of role predictability and word predictability were addictively observed in the measurement of early processing (i.e., first pass reading times) but not in the measurement of late processing (i.e., regression rates). Of interest, the effect of word predictability was mainly based on the processing of words of low role predictability. Our results indicate that when a thematic role for an upcoming word is not highly expected, word predictability contributes to the reduction of readers` integration difficulty, in a way that more predictable word is easier to be integrated into a sentence. We discuss our results in terms of Staub`s (2015) theoretical conclusion that the predictability effect depends on the process that multiple words are activated at once at graded degrees from the early stage. (Gachon University and Konkuk University)

The Effects of Word Predictability and Contextual Uncertainty in the Processing of Korean Dative Sentences: An Eye-Movement Reading Comprehension Study ¹

( Hongoak Yun ),( Dongsu Lee ),( Upyong Hong ) 서울대학교 인지과학연구소 2017 Journal of Cognitive Science Vol.18 No.2

The purpose of this study was to investigate the conjoined roles of word predictability and contextual uncertainty in modeling readers` processing difficulty in the integration of recipients or patients into Korean dative sentences, by observing readers` eye movements in reading. The recipients and patients were arranged in a canonical order (i.e., recipient before patient) or a non-canonical order (i.e., patient before recipient). Using a cloze task, we measured a word`s conditional probability to estimate the degree of predictability for a target word given a context. The outputs from the clozetype listing task were used to compute entropy-based contextual uncertainty corresponding to the degree of contextual constraint at the point in which target words would appear. While the effect of word predictability on eye movement in reading was being taken into account in linear mixed-effect models, the significant effect of contextual uncertainty emerged only in the processing of target words in non-canonical sentences in a way that words were read more rapidly as the strength of contextual uncertainty increased. Including the contextual-uncertainty factor in the model of word predictability improved the goodness of the model fit significantly in the condition of canonical sentences and marginally significantly in that of non-canonical sentences. Our results suggest that the role of contextual uncertainty is important to improve the model performance accounting for expectation-based probabilistic readers` behaviors during sentence comprehension. A weak context (i.e., high contextual uncertainty) in which a wide distribution of possible choices is constructed contributes to the reduction of processing difficulty, in particular, when readers do not make a strong commitment for upcoming structural information.

Segmental Acoustic Correlates Associated with the Korean Lenis Stops

Hongoak Yun 서울대학교 언어교육원 2013 語學硏究 Vol.49 No.1

The purpose of this study was to investigate which acoustic correlates to stop feature distinction were realized due to the presence of a segmentation boundary and how semantic focus information contributed to acoustic correlates associated with segmental units. A production study was conducted using Korean lenis stops. The closure durations and VOTs of lenis stops were longer when they were in boundary-initial positions than when they were in boundary-internal positions. The VOTs became longer as the position of words in the structural hierarchy of sentences grew higher. The F0s of vowels were lower when they followed boundary-initial lenis stops than when they followed boundary-internal lenis stops. The semantic focus information resulted in longer VOTs of initial lenis stops and lower F0s in vowels following initial lenis stops. These results indicated that talkers reliably indicated the onset of words and phrases in the closure durations and the VOTs of lenis stops and the F0s of their following vowels.

The Role of Frequency in the Processing of giving and receiving Events in Korean

Hongoak Yun,Eunkyung Yi 서울대학교 언어교육원 2019 語學硏究 Vol.55 No.2

This study aimed to examine the processing benefits of frequency information associated with the case marker -eykey in comprehending Korean declarative sentences. By using a picture description task in which pictures ambiguously illustrated either a giving event (-eykeyREC … cwuta ‘give … to’) or a receiving event (-eykeySOURCE … patta ‘receive … from’), we found that giving events were predominantly preferred to receiving events. The results of the online sentence comprehension study revealed that 1) give-type verbs were integrated into sentences faster than receive-type verbs overall and 2) the reading-time differences between the verb types were significant when role NPs were canonically ordered (NP-eykey … NP-(l)ul) but not when they were noncanonically presented (NP-(l)ul … NP-eykey). We claim that structural and semantic frequency bias associated with -eykey facilitates readers’ anticipatory processing in the integration of upcoming information. We further discuss how the processing differences in giving and receiving events might attribute to the argument-adjunct distinction between recipients and sources.

The effect of role shifting and expectation in the processing of center-embedded relative clauses in Korean

( Hongoak Yun ),( Yunju Nam ),( Duck Geun Yoo ),( Upyong Hong ) 경희대학교 언어연구소 2015 언어연구 Vol.32 No.2

The purpose of this study is to examine whether the effect of role shifting and expectation serves an independent function in predicting the degree of processing difficulty in center-embedded relative clauses in Korean. In Experiment 1, we observed that head NPs modified by relative clauses, regardless of the order of constituents (i.e., SORELV or OSRELV), took longer to read when the roles corresponding to the traces of the head NPs needed shifting than when they needed not. In Experiment 2, we found that probabilistic distributions pertaining to the head NPs differed as a function of whether or not role shifting for the NPs was required. The mixed-effect models with expectation playing as a predictor on processing difficulty behaved similarly to the model with role shifting being a predictor on processing difficulty. However, mediation analyses in which expectation and role shifting were considered in the same model yielded that the effect of expectation subsumed that of role shifting when constituents were canonically ordered but not when they were scrambled. We claim that the fundamental function of expectation in association with role shifting is additionally effective only when sentence complexity is not extremely severe. (Konkuk University ㆍ Hankuk University of Foreign Studies)

Individual Differences in the Expectation-based Comprehension: An ERP Study

Hongoak Yun,Dongsu Lee,남윤주 서울대학교 인지과학연구소 2022 Journal of Cognitive Science Vol.23 No.1

Readers’ active use of linguistic cues from a given context elicits anticipatory processing of yet-to-be-encountered information. In this study, we aimed to examine whether the patterns of anticipatory comprehension would systematically differ by the degree of readers’ working memory capacity. Readers’ evoked responses potentials (ERPs) in response to words were recorded in the processing of Korean dative sentences (i.e., subject+[recipient+theme]/[theme-recipient]+adverb+verb) in which the presentation order of arguments (i.e., role predictability) and the likelihood corresponding to argument role fillers (i.e., word predictability) were manipulated. We found quantitative and qualitative differences in ERPs among readers during sentence comprehension. The N400 emerged in the integration of unpredictable words, and it occurred more frequently among readers with low working memory. Of our interest, we observed the asymmetrical distribution of the negativity and the positivity, attributable to the differences in readers’ working memory, at adverbs and verbs in which readers were busy with integrating previously-presented arguments into sentences and processing incoming words. Our results suggested that readers with low working memory are more involved in the lexical retrieval process, whereas those with high working memory are more attentive to the structural or semantic integration process. In short, we argued that the lack of working memory capacity could make readers fall behind in using lexical and structural information during sentence processing, particularly for argument integration.

Predictive Representation for Upcoming Linguistic Input in Younger and Older Adults

( Hongoak Yun ),( Sun Hyun Moon ),( Soo Rim Noh ) 한국언어정보학회 2018 언어와 정보 Vol.22 No.2

The immediate use of the contextual information to predict the upcoming information during sentence comprehension may vary depending on the aspects of the linguistic and world knowledge and the cognitive ability of readers. In this study, we investigated how well older readers employ their linguistic and world knowledge to develop the expectation for the upcoming information during sentence comprehension. In two offline tasks (cloze and listing), older and young readers were asked to complete sentence fragments (when agents and recipients were given or when agents and patients were given). The responses were used to compute predictability measurements (structural and lexical predictability). We found no aging effect in the production of likely structural choices and likely lexical choices, meaning that the most predictable ones to older readers were also the most predictable to young readers. Significant aging effect emerged in the production of unlikely choices; older readers were more predictable to unlikely structures but less predictable to unlikely words. We discuss the contextual sensitivity of older readers in formulating their knowledge representation for what is coming up next.

Exploring AI-Generated English Relative Clauses in Comparison to Human Production

Hongoak Yun,이은경,송상헌 서울대학교 인지과학연구소 2023 Journal of Cognitive Science Vol.24 No.4

Human behavioral studies have consistently indicated a preference for subject-extracted relative clauses (SRCs) over object-extracted relative clauses (ORCs) in sentence production and comprehension. Some studies have further shown that this preference can be influenced by the semantic properties of head nouns, particularly animacy. In this study, we use AI language models, specifically GPT-2 and ChatGPT 3.5, to simulate human sentence generation. Our primary goal is to evaluate the extent to which these language models replicate human behaviors in sentence production and identify any divergences. We tasked the models with completing sentence fragments structured as ‘the,’ followed by a head noun and 'that’ (The reporter that …). We varied the semantic property of head nouns such that they are all animate (the secretary that … ) in Study 1 and are either animate or inanimate (the musician/book that … ) in Study 2. Our findings reveal that in Study 1, both GPT models exhibited a robust SRC bias, replicating human-like behavior in relative clause production. However, in Study 2, we observed divergent behavior between the models when head nouns were inanimate, while consistency was maintained when head nouns were animate. Specifically, ChatGTP 3.5 generated more ORCs than SRCs in the presence of inanimate head nouns. These results, particularly those from ChatGPT 3.5, closely mirror human relative clause production patterns. Our study highlights the potential of language generative language models as efficient and versatile corpus simulators. Furthermore, our findings contribute to the evolving field of AI linguistics, shedding light on the capacity of AI generative systems to emulate human-like linguistic patterns in sentence production.

The Effect of anticipation in the ambiguity resolution of -eykey during sentence comprehension

Hongoak Yun,Eunkyung Yi 미래영어영문학회 2019 미래영어영문학회 학술대회 자료집 Vol.2019 No.6

Individual Differences in Expectation-based Sentence Comprehension

Hongoak Yun 미래영어영문학회 2018 미래영어영문학회 학술대회 자료집 Vol.2018 No.6