The evidence behind the method.

Japanese speakers could not reliably distinguish /l/ from /r/ in perception tests, performing near chance, while American English speakers were near-perfect. Established the difficulty as rooted in the absence of a phonemic contrast in Japanese — a problem that begins before speech is even attempted, not a motor-production deficit.

Proposed that learners fail to acquire a new phoneme when the target is close enough to a native sound that the brain maps it onto the existing category. For Japanese speakers both /l/ and /r/ map onto the single Japanese liquid, making the distinction extremely resistant to change without explicit perceptual training.

Japanese speakers trained on /l/–/r/ with high-variability phonetic training improved in both perception and production. Confirmed the link between perceptual failure and production error — those who could not hear the difference could not produce it — validating the substitution as a measurable, trainable deficit.

Japanese speakers systematically "hallucinate" vowels when perceiving consonant clusters, reporting an extra syllable that isn't in the signal. This perceptual illusion directly predicts epenthesis in production: speakers insert vowels because their phonological system represents the word with them.

Epenthetic vowel insertion is not random but follows rules tied to the mora structure of Japanese; inserted vowels match the quality Japanese phonotactics expect. Confirms the L1 syllable template being imposed onto L2 words.

Even highly proficient Japanese speakers continue to show epenthesis, particularly at word boundaries and in initial clusters. Its persistence in advanced learners indicates it needs explicit targeted training rather than resolving through exposure alone.

Japanese learners transfer mora-timed rhythm from their L1, producing syllables of roughly equal length and prominence instead of English's alternating pattern — directly causing lexical stress errors because the acoustic cues for stress are flattened.

Of the features that reduced intelligibility for English listeners, lexical stress errors had a disproportionately large effect on comprehension — larger than vowel or consonant substitutions. Misplaced stress on a specialised term can stop a listener recognising the word at all.

Compared Japanese and English prosody directly, showing Japanese pitch accent and mora timing create a fundamentally different mental representation of word rhythm. Japanese speakers are operating from a different underlying system, not simply producing stress incorrectly.

Filler rate is directly correlated with lexical-retrieval difficulty. For Japanese speakers in technical English the effect compounds — both general L2 access difficulty and domain vocabulary load — predicting systematically higher filler rates than native speakers on the same tasks.

Japanese speakers used significantly more filled pauses in English than in Japanese, and more than native English speakers on equivalent tasks. The increase was largest in informationally dense speech — precisely a technical explanation task.

High filler rates significantly reduced listener ratings of fluency, confidence, and expertise even when content was identical — directly relevant to professional settings where perceived competence affects collaboration, promotion, and trust independent of actual ability.

Established the link between silent pauses and cognitive planning difficulty — pauses cluster before semantically complex, low-predictability words. Japanese speakers in technical English face planning bottlenecks that surface as mid-clause pauses at a measurably higher rate.

Non-native speakers placed a significantly higher proportion of pauses within clauses rather than at boundaries (which native speakers rarely do). Mid-clause pausing was identified as the most reliable acoustic marker separating fluent from disfluent L2 speech.

More cognitively demanding tasks — technical explanations with cause-and-effect reasoning — produced disproportionately more mid-clause pauses in non-native speakers, confirming that a technical speaking task specifically elicits the behaviour this challenge targets.

Developed the self-repair framework: an internal monitor detects an error or planning failure and interrupts, leading to a cutoff and restart. L2 speakers trigger this monitor far more often because they manage planning, retrieval, and encoding in a less automatised system.

Applied the self-repair model to L2 speakers: non-natives produced significantly more repetitions and restarts, triggered most often by lexical-search failure. Japanese speakers of English showed a particularly high rate of mid-utterance restarts on content words.

Tasks requiring speakers to organise and explain information under time pressure produced the highest disfluency rates — confirming repetitions and false starts as a valid, task-sensitive measure of L2 fluency rather than a stylistic trait, supporting their use as a pre/post metric.