When Grammar Rules Stick Too Well: How L2 Learners Overgeneralize Japanese Sound-Symbolic Verbs

Japanese sound-symbolic words, known as onomatope, are everywhere in the language, from manga dialogue to everyday conversation. Yet for second-language learners, these vivid expressions are notoriously difficult to master. A new study published in Frontiers in Psychology reveals that Chinese-speaking learners of Japanese systematically overextend one grammatical rule—noun phrase attachment—to contexts where Japanese grammar forbids it, and that only the most proficient learners manage to break the habit.

The research, conducted by Yajing Feng of Hangzhou Normal University and Katsuo Tamaoka of Shanghai University and Nagoya University, focused on a subclass of Japanese mimetics called psychomimes, which describe physical and psychological sensations. Unlike most languages, Japanese divides its sound-symbolic vocabulary into three categories: phonomimes that imitate sounds, phenomimes that depict visual and tactile events, and psychomimes that encode internal bodily and emotional states. Psychomimes are the hardest of the three for learners because their connection to actual sounds is weak, forcing learners to depend on abstract grammatical structures to interpret their meaning.

The grammatical structure in question is noun phrase (NP) attachment. Japanese psychomime verbs built with the light verb suru fall into three classes. Class 1 words, such as zukizuki (throbbing pain), describe involuntary bodily sensations and require an attached noun phrase naming the affected body part, as in atama-ga zukizuki shite iru (the head is throbbing). Class 2 words, such as wakuwaku (feeling excited), convey emotions through physical reactions and allow the noun phrase optionally. Class 3 words, such as kuyokuyo (feeling gloomy), express purely psychological states and reject the noun phrase altogether. The authors drew on Kageyama’s lexical conceptual structure framework, in which verbs whose central meaning involves movement or change mark an internal argument, while verbs of experience or control mark an external argument, to explain why these syntactic differences arise.

To test how learners handle these distinctions, the researchers ran two paper-and-pencil experiments using a noun phrase attachment judgement task. In the first experiment, 76 Japanese native university students judged 18 psychomime words, six from each class, choosing among a matching noun phrase, an unrelated distractor, or no noun phrase at all. None of the 18 words appeared in the Japanese Language Proficiency Test vocabulary lists or in the learners’ textbooks, and post hoc correlation analyses confirmed that word frequency did not systematically bias the judgments. The test showed acceptable internal consistency, with a Cronbach’s alpha of 0.76.

The native speaker results validated the three-class division. Class 1 items drew noun phrase attachment at nearly 100 percent, a clear ceiling effect. Class 2 items attracted attachment at a mean rate of 75.44 percent, significantly above chance, though the individual items formed a gradient: mukamuka and kurakura reached about 91 percent, while uzuuzu actually fell below chance. Class 3 items showed a mean suppression rate of 72.37 percent, with four of the six words significantly favoring suppression and the boundary-blurring words gutaguta and piripiri showing no significant preference. A repeated-measures analysis of variance confirmed the ordering Class 1 greater than Class 2, which differed from Class 3, with a large effect size, supporting the three-category classification over an earlier two-category proposal.

The second experiment turned to 88 Chinese undergraduate Japanese majors, none of whom had studied in Japan. Participants completed the same judgement task along with a validated 36-item L2 grammar test covering morphological inflections, local dependency, and complex structures. Based on their grammar scores, learners were divided into low, middle, and high proficiency groups, and the grouping was verified with a one-way analysis of variance showing a large effect. Only about 7 percent of responses were distractor selections, indicating that comprehension of the test sentences was not a major obstacle.

The learner results revealed a strikingly category-specific pattern of overgeneralization. On Class 1 items, where Mandarin and Japanese both require the noun phrase, learners performed well, with a mean attachment rate of 73.77 percent, significantly above chance. But on Class 3 items, where Japanese requires suppression, learners attached noun phrases at 56.46 percent, and two items, gutaguta and uziuzi, drew significantly more attachment than suppression choices. In other words, learners mechanically extended the attachment rule from contexts where it belongs to contexts where it is grammatically illicit. The authors interpret this through Markedness Theory: learners overgeneralize rules from unmarked structures to marked ones, especially when the marked target structure corresponds to an unmarked pattern in the native language.

Mandarin’s properties help explain the pattern. Chinese bodily-sensation expressions, like wo de tou teng (my head aches), require the body-part noun, matching Japanese Class 1 behavior, so no cross-linguistic conflict arises there. But Chinese psychological-state expressions allow the heart noun optionally, as in either wo hen fennu (I am very angry) or wo xinli hen fennu (I am angry in my heart). Lacking a consistent attachment-versus-suppression contrast, and lacking Japanese case-marking cues such as the ga particle, learners default to attachment as a general strategy. Crucially, however, this reliance proved dynamic rather than permanent. A classification tree analysis showed that psychomime type was the dominant factor predicting learners’ choices, with proficiency emerging as a significant secondary factor for Classes 2 and 3: only high-proficiency learners suppressed attachment at above-chance rates, while middle- and low-proficiency learners continued to overgeneralize, particularly on Class 3 items.

Two further analyses sharpened the picture. A multiple linear regression found that both Class 1 and Class 3 accuracy independently predicted grammar proficiency, though together they explained only about 13 percent of its variance. A binary logistic regression showed that Class 2 performance was the strongest predictor of learners’ overall attachment preference, with an odds ratio of 4.82, exceeding those of Class 1 and Class 3. Because Class 2 sits at an intermediate point on the continuum from concrete bodily sensation to abstract psychological state, requiring learners to integrate both physical and psychological dimensions of meaning, it imposes the greatest cognitive demands and serves as the most sensitive indicator of developing syntactic competence. Notably, proficiency entered as a continuous variable did not significantly predict attachment preference, confirming that its influence is non-linear: a threshold effect, rather than a gradual improvement with each increment of grammatical knowledge.

The authors frame these findings within U-shaped developmental models and Processability Theory. Suppressing a noun phrase, they argue, is a more complex grammatical operation than producing one, so learners master attachment first and only later, once sufficient processing capacity and analytical ability develop, learn to suppress it. Overgeneralization, in this view, is not a sign of failure but a natural stage in restructuring an interlanguage grammar. The study also carries pedagogical implications: explicit contrastive instruction between psychomime classes, Class 2 items as diagnostic tools, and corrective feedback timed for learners who have already crossed the developmental threshold may all help learners differentiate the rules. The authors caution that their participants came from a single university in each country, that only Chinese-speaking learners were tested, and that the cross-sectional, paper-based design cannot capture how individual learners’ biases change over time, leaving longitudinal and cross-linguistic follow-up work as the natural next step.

Subject of Research: Overgeneralization of noun phrase attachment rules in second-language acquisition of Japanese psychomime sound-symbolic verbs

Article Title: When rules stick too well: overgeneralization of physical and psychological noun phrase attachment in L2 learning of Japanese sound-symbolic words

Article References: Feng, Y., & Tamaoka, K. (2026). When rules stick too well: overgeneralization of physical and psychological noun phrase attachment in L2 learning of Japanese sound-symbolic words. Frontiers in Psychology, 17, Article 1906329. https://doi.org/10.3389/fpsyg.2026.1906329

Image Credits: AI Generated

DOI: 10.3389/fpsyg.2026.1906329

Keywords: Japanese sound-symbolic words, psychomimes, noun phrase attachment, overgeneralization, second language acquisition, L1 transfer, grammatical proficiency, syntax-semantics interface, Markedness Theory, U-shaped development, Chinese learners of Japanese, threshold effect