探討中日同形詞語義重合方式對日中雙語者辨識中文詞彙的影響

The Effect of Sino-Japanese Cognates and Homographs with Different Degrees of Semantic Overlap on Chinese Lexical Decision by Japanese Learners of Chinese as a Second Language

本研究以65位日中雙語者為研究對象，探討五種語義重合方式的中日同形詞（同形同義、日＜中、日＞中、日＞＜中、同形異義）如何影響其從事中文詞彙判斷的表現。結果顯示，日中雙語者在預期純日文詞可能出現的心理下，同形同義詞與控制詞相比也失去促進作用，但與其他同形詞相比，在正確率和反應時間上仍佔優勢。同形異義詞和日＞＜中詞的反應時間比同形同義詞長，表示語義不重合或僅少部分重合且在中、日文語意範圍都廣的詞，會產生較大的抑制效應。進一步考量日中雙語者的華語程度、對詞彙用法熟悉度與同形詞音似度的效應時發現，三者對反應正確率都有正向的影響。華語較低程度（B1）者在日＞中和同形異義詞正確率最低，顯示同形詞在日文語義較廣或語義完全不同都容易造成認知衝突。此外，中日文詞頻的效果不一致，中文詞頻有促進反應的正向作用，但日文詞頻卻有負向的抑制作用，且詞頻與其他因子也有顯著的交互作用。整體結果指向日中雙語者在從事中文詞彙判斷任務時，仍會非選擇性地激活日文詞彙的意義，而隨著同形詞語義重合部分減少或語義範圍更廣，反應時間增加，正確率降低。此結果支持同形詞雙詞素表徵並符合BIA＋模型的預測，即字形、語音、語義協調共振可產生促進效應。

This study investigates how different patterns of semantic overlap in Sino-Japanese cognates influence Chinese lexical decision performance among Japanese learners of Chinese as a second language (CSL). While previous research on bilingual lexical access has primarily focused on Indo-European language pairs and has often adopted a binary distinction between cognates and interlingual homographs, the present study addresses a critical gap by examining the fine-grained semantic diversity of cognates in a non-alphabetic writing system. Specifically, it explores how varying degrees and configurations of semantic overlap between Chinese and Japanese cognates modulate response accuracy and response time in a Chinese lexical decision task, and how these effects interact with learners' Chinese proficiency, lexical familiarity, phonological similarity, and word frequency in both languages. A total of 65 Japanese-Chinese bilinguals and 45 Chinese monolingual controls participated in the experiment. The bilingual participants were native speakers of Japanese learning Chinese in Taiwan and ranged from intermediate (B1) to advanced (C1) proficiency levels, as determined by HSK or TOCFL speedy screening tests. The monolingual control group consisted of native Mandarin speakers with no knowledge of Japanese. All participants completed a computerized Chinese lexical decision task in which they were required to judge whether visually presented two-character items were valid Chinese words. Importantly, bilingual participants were informed in advance that pure Japanese words written in Chinese characters could appear and that these should be rejected, thereby inducing an expectation of cross-language interference. The experimental stimuli consisted exclusively of two-character Chinese words and nonwords and included five categories of Sino-Japanese cognates classified according to their semantic overlap patterns: (1) identical cognates with fully overlapping meanings in both languages, (2) Japanese < Chinese cognates, in which the Japanese meaning constitutes a subset of the Chinese meaning, (3) Japanese > Chinese cognates, in which the Japanese meaning is broader than the Chinese meaning, (4) Japanese > < Chinese cognates, which share only partial semantic overlap while also having language-specific meanings, and (5) interlingual homographs with entirely non-overlapping meanings. Two types of non-cognate control words were also included: pure Chinese words and pure Japanese words. In addition, Chinese-like nonwords were used as fillers. After the lexical decision task, bilingual participants rated a subset of the stimuli for usage familiarity and phonological similarity between Chinese and Japanese readings. Response accuracy was analyzed using generalized linear mixed-effects models, while response times for correct responses were analyzed using linear mixed-effects models after appropriate transformation. Both subject-level and item-level random effects were included to ensure generalizability beyond the sampled participants and materials. Signal detection measures (d' and response bias c) were additionally computed to verify that bilingual participants were not responding by random guessing or by adopting a systematic response bias. The results demonstrate several robust findings. First, when bilingual participants anticipated the possible appearance of Japanese words, identical cognates no longer exhibited a facilitative effect relative to pure Chinese control words. Instead, their response times were significantly longer than those for pure Chinese words, indicating that even fully overlapping cognates can induce inhibitory effects when cross-language competition is heightened by task demands and expectancy. Nevertheless, identical cognates still showed clear advantages in both accuracy and response time compared to other cognate and homograph types, confirming that semantic overlap continues to play a mitigating role under conditions of non-selective activation. Second, semantic overlap patterns significantly modulated bilingual lexical decision performance. Cognates and homographs with partial or no semantic overlap elicited lower accuracy and longer response times than identical cognates. In particular, Japanese > < Chinese cognates and interlingual homographs produced the strongest inhibitory effects, suggesting that items with broad and divergent semantic domains across languages generate greater cognitive conflict. These results indicate that semantic competition is not simply a function of whether meanings overlap, but also of how semantic domains are structured and distributed across the two languages. Third, bilingual participants showed clear evidence of non-selective lexical access. Even in a task requiring responses only to Chinese words, Japanese lexical representations were automatically activated, as evidenced by reduced accuracy and prolonged response times for cognates and by the difficulty participants experienced when rejecting pure Japanese words. Importantly, rejecting Japanese words was significantly slower and less accurate than rejecting nonwords, confirming that native-language lexical status alone is sufficient to induce interference, even when participants are explicitly instructed to inhibit it. Fourth, individual difference factors significantly shaped bilingual performance. Higher Chinese proficiency was associated with higher accuracy across all cognate types, with particularly large gains observed for semantically mismatched items such as Japanese > Chinese cognates and interlingual homographs. This suggests that increased L2 proficiency enhances learners' ability to resolve cross-language semantic conflict. Lexical familiarity also emerged as a strong predictor: Words rated as more familiar were processed more accurately and more quickly, indicating that experiential knowledge can partially compensate for semantic ambiguity. Phonological similarity, while not exerting a strong main effect on response time, significantly improved response accuracy, especially for low-frequency Chinese words. This finding suggests that phonological information may play a supportive role at later decision stages. Such effects are noteworthy given the relatively opaque phonological mapping of Chinese characters and the variability of Japanese readings, and they point to a more nuanced role of phonology in character-based bilingual processing than previously assumed. Word frequency effects further revealed an asymmetric influence of the two languages. Higher Chinese word frequency facilitated both accuracy and speed, whereas higher Japanese word frequency exerted an inhibitory effect on accuracy, particularly for cognates with broad or partially overlapping semantic domains. These findings indicate that L1 word frequency can amplify cross-language interference, while target-language word frequency strengthens lexical access and decision confidence. Moreover, interactions between word frequency and proficiency showed that advanced learners benefited more from high Chinese word frequency than intermediate learners, suggesting that frequency effects are increasingly exploited as lexical representations become more entrenched. Notably, Chinese monolingual participants did not exhibit systematic differences across cognate categories once lexical familiarity was controlled, confirming that the observed effects among bilinguals were genuinely attributable to cross-language activation rather than intrinsic lexical properties of the stimuli. Taken together, the findings provide strong support for non-selective bilingual lexical access and align closely with the predictions of the Bilingual Interactive Activation Plus (BIA+) model and related computational frameworks. The results are also consistent with the notion of dual-lexical representations for cognates, even when orthographic forms are identical. Orthographic overlap alone is sufficient to activate representations from both languages, while semantic overlap determines whether this activation results in facilitation or inhibition. Furthermore, the study extends existing models by demonstrating how semantic domain breadth, L2 proficiency, lexical familiarity, phonological similarity, and asymmetric word frequency jointly shape bilingual lexical decision processes in a non-alphabetic writing system. In conclusion, this study advances the understanding of bilingual lexical processing by offering a fine-grained analysis of semantic overlap in Sino-Japanese cognates. It demonstrates that semantic overlap is not a binary property but a multidimensional construct with measurable cognitive consequences. By integrating linguistic, cognitive, and experiential factors, the study provides new evidence for the universality of non-selective lexical activation while highlighting languagespecific characteristics of character-based bilingualism. These findings have important implications for theories of bilingual lexical access as well as for Chinese language pedagogy, particularly in the instruction of Sino-Japanese cognates that are prone to semantic divergence and learner confusion.