Decoding the TOEIC Link Speaking Pronunciation Rubric — The 5 Axes ETS Scorers Watch and How to Self-Train Each One

Pronunciation scoring is the sum of five axes, not "native-like" impression

Test-takers stuck on the pronunciation sub-score almost always train against the wrong target — "speak more like a native". That target is invalid because ETS scorers (and the ETS automated scoring engine) do not score impression. They score five independent observation axes and sum the result.

The five axes are (1) segmental accuracy, (2) word stress, (3) rhythm and function-word reduction, (4) vowel length and consonant deletion, (5) intonation contour. Because each axis adds independently, fixing the weakest one or two axes can lift the pronunciation score by 10-15 points without touching the others.

• Scorers accumulate per-axis observations, not impression
• Fix the weakest 1-2 axes first — that is the shortest path
• L1 Japanese test-takers concentrate deductions on axis (3) and (4)
• Axis (1) is most efficiently trained against L1 minimal pairs

Axis 1: Segmental accuracy

The first axis to train. The scorer is checking whether you can hear and produce the sound contrasts that do not exist in Japanese (/r/ /l/ /θ/ /ð/ /æ/ /v/ /f/ /w/) and the contrasts that map confusingly (/s/ /ʃ/, /i:/ /ɪ/).

The drill is mechanical. Build a list of 50 minimal pairs, record yourself reading them, compare your formant frequencies against a reference recording in Audacity (free) or Praat (free). Visual formant comparison makes it visible that your /æ/ is drifting into /ʌ/, which you cannot hear.

• Top 5 pairs to drill: rice/lice, bath/bus, sheep/ship, very/berry, think/sink
• Build a 50-pair list, drill 10 minutes a day
• Run weekly formant comparison in Audacity or Praat
• Use phone speech-to-text and check whether homophones get misrecognised

Axis 2: Word stress

English multisyllabic words have fixed stress placement. Misplaced stress sounds like a different word — "PHOtograph" / "phoTOgraphy" / "photoGRAPHic" all use a different stress slot. Reading aloud without checking stress placement produces a 50%-plus error rate on multisyllabic words.

Scorers tend to penalise stress errors as roughly two segmental errors, because a stress error degrades comprehension of the whole sentence. The deduction per error is high.

• Always check stress placement for new words before shadowing
• For 3+ syllable words, mark the stressed syllable in capitals
• Same-form noun/verb pairs (record/record, conduct/conduct) flip part of speech via stress
• Learn 5 stress-shift rules for common suffixes (-tion / -ity / -ic)

Axis 3: Rhythm — function-word reduction

The single highest-deduction axis for L1 Japanese test-takers. English is stress-timed: content words (verbs, nouns, adjectives, adverbs) are stressed; function words (prepositions, auxiliaries, articles, pronouns) are reduced. Japanese syllable-timed reading produces equal weight on every word and the rhythm collapses.

The fix is direct training on weak forms and contractions. "to" reduces to /tə/, "and" to /ən/, "have" to /əv/, "could have" to /ˈkʊdəv/. Test-takers who cannot hear these reductions in textbook audio will not produce them either.

• Memorise 10 weak-form pairs: to/of/and/the/a/at/for/from/can/have
• Memorise 5 contractions: would have → wouldʼve, going to → gonna (informal)
• Read aloud one clause at a time consciously stressing content words
• Practise with a metronome at 60-72 BPM, content words on the beat

Axis 4: Vowel length and consonant deletion

English vowels carry phonemic length. Long vowels /i:/ /u:/ /ɔ:/ /ɑ:/ are physically 1.5-2x longer than short vowels. L1 Japanese speakers tend to lengthen short vowels (sit → seat), which scores as a vowel-length deduction separate from the segmental error.

Consonant cluster deletion matters too. "asked" is /ɑ:skt/ — four phonemes in one beat — but Japanese speakers tend to insert vowels and produce four beats /a-su-ku-do/. Drill word-final clusters -ed, -ts, -sk to keep them as a single beat.

• Lengthen long vowels (sheep / pool / saw / car) to ~2x duration
• Cut short vowels (ship / pull / sock / cat) cleanly
• Compress word-final -ed / -ts / -sk / -lk into one beat
• 5 minutes a day on "asked", "tests", "milk" as one-beat words

Axis 5: Intonation contour

The axis you discriminate rising vs falling pitch on sentence boundaries. Yes/no questions and list-medial items rise (↗); statements, wh-questions and list-final items fall (↘). Marking ↗/↘ on a transcript before reading aloud surfaces your own intonation awareness.

Topic-boundary pitch reset matters too. Reading English with flat Japanese-style prosody erases sentence boundaries for the scorer, and the comprehension assessment for the whole response drops.

• Yes/No question → end on rise (↗)
• Statement, wh-question, command → end on fall (↘)
• Lists go "A↗, B↗, and C↘"
• Reset pitch one step lower at topic boundaries

Axis training cost vs deduction risk

* Deductions concentrate on axes 3 and 4 because L1 Japanese is syllable-timed with isochronous mora; axes 1-2 are dictionary-checkable per word, but axes 3-4 require habit rewriting and take longer.

Frequently Asked Questions