In a groundbreaking study from Waseda University, researchers have shed new light on how Japanese learners of English process ambiguous sentences in real time. Led by Associate Professor Chie Nakamura from the School of International Liberal Studies, the team explored whether second-language (L2) learners rely on the same predictive mechanisms as native speakers or if they favor different cues. The findings, published in Frontiers in Language Sciences on March 4, 2026, reveal a gradient approach to sentence processing, where proficiency in structural computation determines whether learners use syntax-driven predictions or fall back on lexical associations.

This research is particularly relevant amid Japan's ongoing efforts to boost English proficiency. Despite reforms like increased English-taught programs at top universities and government initiatives to enhance communication skills, Japan ranked 96th out of 123 countries in the 2025 EF English Proficiency Index, falling into the 'very low proficiency' category for the first time. With over 90% of Japanese students studying English from elementary school, understanding real-time comprehension challenges can transform teaching strategies.

🧠 The Foundations of Predictive Language Processing

Human language comprehension is remarkably efficient—we predict upcoming words and structures before hearing them fully. This predictive processing relies on cues like verb subcategorization (e.g., 'tell' expects a complement) and syntactic hierarchies. In native speakers, eye-tracking studies using the visual-world paradigm show anticipatory looks to likely referents, reflecting active parsing.

Filler-gap dependencies exemplify this: in wh-questions like "Where did Lizzie tell someone that she was going to catch butterflies?", the wh-word ('where') seeks a gap. Native English speakers predict the matrix clause gap after 'tell', fixating on the 'telling' scene despite ambiguity until the embedded clause disambiguates. Japanese, a head-final language with in-situ wh-phrases, processes dependencies differently, often verb-driven rather than filler-driven.

Prior L2 research debated if non-natives predict like natives or use 'shallow' processing, relying on lexical semantics over syntax. This Waseda study bridges the gap by examining Japanese-English learners on globally ambiguous filler-gaps.

🔬 Methods: Eye-Tracking the Invisible Parse

The experiment involved 50 upper-intermediate to advanced Japanese L2 learners (CEFR B2-C1, 6+ years English study, no immersion) and 25 native English speakers. Participants heard context sentences describing two events (matrix clause: telling location; embedded clause: action location), viewed pictures of both plus a distractor, then an ambiguous wh-question.

Eye movements (1000Hz EyeLink) tracked fixations from question onset. Learners were split by accuracy on unambiguous fillers (high: 83% matrix preference; low: 55%). Permutation analyses identified significant look clusters to matrix vs. embedded pictures.

Stimuli balanced event order to control recency. Key-press responses measured final interpretation.

📊 Key Findings: A Tale of Two Predictions

Natives showed a late, sustained matrix bias post-matrix verb, structure-driven. All L2 learners predicted early via lexical cues ('where-tell' association), then shifted to structure. Crucially, high-accuracy learners mirrored natives—no early lexical bias, robust structural prediction. Low-accuracy showed early lexical looks, fragmented later structure use, and no reliable final preference.

• High-Accuracy L2: Native-like timing, structure-guided.
• Low-Accuracy L2: Lexical-driven early, unstable structure later.
• No L1 transfer: Learners adopted English active-filling, not Japanese verb-waiting.

View the detailed results in the open-access paper: Lexical vs. structural cue use in L2 prediction.

🌉 Bridging Native and L2 Processing: The Gradient Model

Challenging binary views (native-like vs. shallow), the study supports a gradient: cue weighting shifts with syntactic proficiency. High-ability learners recruit structure rapidly; low-ability lean on frequent lexical pairs but struggle integrating syntax under time pressure.

This aligns with convergence hypotheses—L2ers approximate L1 mechanisms but vary individually. No 'good enough' processing universally; instead, ability modulates cues.

🎓 Implications for Japanese English Education

Japan's English challenges stem partly from rigid grammar-translation methods, ignoring real-time prediction. With 51.6% of high schoolers at Pre-2 Eiken (CEFR A2) in 2025, up slightly but still low, predictive skills lag.

Teachers should emphasize listening to natural input, filler-gap drills (e.g., wh-questions), and feedback on predictions. Waseda-like research informs curricula, as universities shift to English programs (e.g., UTokyo engineering fully English by 2026).

Explore Waseda press release for quotes: Uncovering Structural Cues.

🏛️ Waseda University's Linguistics Legacy

Waseda, with its School of International Liberal Studies, leads SLA psycholinguistics. Nakamura's work builds on her eye-tracking expertise (e.g., prosody-syntax integration, CELER corpus). Institute of Language and Speech Science fosters interdisciplinary research.

In Japan, where English is mandatory yet proficiency stalls, such studies drive reforms toward communicative competence.

🔮 Future Directions and Broader Impacts

Longitudinal tracking of cue shifts, neuroimaging integration, AI models mimicking L2 prediction. Applications: speech tech for learners, noisy-environment training.

For Japanese higher ed, enhances global competitiveness amid 'Global 30' expansions.

Conclusion: Reshaping L2 Comprehension Paradigms

Waseda's study redefines L2 processing as ability-dependent cue dynamics, offering actionable insights for Japan's English challenge. By prioritizing structure-building, educators can unlock fluent comprehension.

Photo by Jerry Wang on Unsplash