Complex elision and assimilation
A B2 learner knows what did you means in slow speech. A C1 learner has to recognize did you when an American actually says it: /dɪdʒə/, didja, sometimes just /dʒə/. American fast speech aggressively deletes sounds (elision), shifts sound places to match neighbors (assimilation), and merges adjacent sounds into a single new sound (coalescence). At conversational speeds the citation form did you eat yet? becomes /dʒit‿ʃɛt/ — three syllables, six sounds, and not a single one of the original written sounds is preserved unchanged. Without explicit training, Russian L1 listeners often cannot decode this and feel that “Americans speak too fast.”
This lesson covers four reduction mechanisms: elision of /t/, /d/, /h/ in clusters; assimilation of place; voice assimilation (the “hafta” family); and coalescence (the “didja” family). Each has rules. None is sloppy speech — these are systematic phonological processes that operate on every sentence Americans produce above slow conversational speed.
The four reduction mechanisms
- Elision — full deletion of a sound, especially /t/, /d/, /h/ in specific phonological contexts.
- Place assimilation — a sound shifts its place of articulation to match the following sound.
- Voice assimilation — a sound shifts its voicing to match the following sound.
- Coalescence — two adjacent sounds merge into a single new sound (often /tj/ → /tʃ/, /dj/ → /dʒ/).
These operate productively in fast speech. Knowing the rules lets you both decode incoming speech and produce native-like reductions yourself.
1. Elision of /t/, /d/, /h/
/t/ deletion in clusters
When /t/ appears between two consonants (or at the end of a final cluster before a consonant), it is typically deleted in fast speech.
| Phrase | Citation | Fast |
|---|---|---|
| last week | /læst wik/ | /læs wik/ |
| best friend | /bɛst frɛnd/ | /bɛs frɛnd/ |
| first place | /fɝst pleɪs/ | /fɝs pleɪs/ |
| next day | /nɛkst deɪ/ | /nɛks deɪ/ |
| just give me | /dʒəst ɡɪv mi/ | /dʒəs ɡɪv mi/ |
| asked them | /æskt ðəm/ | /æs(k)t̬əm/ — often /æstəm/ |
/d/ deletion in clusters
Same rule as /t/. Word-final /d/ in a cluster before a following consonant is deleted.
| Phrase | Citation | Fast |
|---|---|---|
| old man | /oʊld mæn/ | /oʊl mæn/ |
| send me | /sɛnd mi/ | /sɛn mi/ |
| cold weather | /koʊld ˈwɛðɚ/ | /koʊl ˈwɛðɚ/ |
| world peace | /wɝld pis/ | /wɝl pis/ |
| friend group | /frɛnd ɡrup/ | /frɛn ɡrup/ |
/h/ deletion in unstressed function words
The /h/ in unstressed he, him, his, her, have, has, had is regularly deleted in connected speech.
| Phrase | Citation | Fast |
|---|---|---|
| I told her | /aɪ toʊld hɝ/ | /aɪ toʊl‿dɝ/ |
| give him | /ɡɪv hɪm/ | /ɡɪv‿ɪm/ |
| should have | /ʃʊd hæv/ | /ʃʊɾəv/ — “shoulda” or “should of” |
| would have | /wʊd hæv/ | /wʊɾəv/ |
| could have | /kʊd hæv/ | /kʊɾəv/ |
| does he | /dʌz hi/ | /dʌz‿i/ |
| where has | /wɛr hæz/ | /wɛr‿əz/ |
The famous “should of / could of / would of” misspelling comes from this elision — natives genuinely hear /ʃʊɾəv/ as shoulda and write it phonetically as “should of.”
Russian L1 problem
Russian retains every consonant. A Russian speaker pronounces every /t/, /d/, /h/ in I should have told her. Native AmE: /aɪ ʃʊɾə toʊl‿dɝ/. The Russian production sounds slow and overly precise.
2. Place assimilation
A consonant shifts its place of articulation to match the following consonant. The most common patterns:
Alveolar → bilabial before bilabial
Final /n/, /t/, /d/ before /p/, /b/, /m/ shift to bilabial /m/, /p/, /b/.
| Phrase | Citation | Assimilated |
|---|---|---|
| good boy | /ɡʊd bɔɪ/ | /ɡʊb bɔɪ/ — /d/ → /b/ |
| ten pounds | /tɛn paʊndz/ | /tɛm paʊndz/ — /n/ → /m/ |
| in May | /ɪn meɪ/ | /ɪm meɪ/ |
| green party | /ɡrin ˈpɑrti/ | /ɡrim ˈpɑrti/ |
| great party | /ɡreɪt ˈpɑrti/ | /ɡreɪp ˈpɑrti/ |
Alveolar → velar before velar
Final /n/, /t/, /d/ before /k/, /g/ shift to velar /ŋ/, /k/, /g/.
| Phrase | Citation | Assimilated |
|---|---|---|
| in case | /ɪn keɪs/ | /ɪŋ keɪs/ |
| ten guys | /tɛn ɡaɪz/ | /tɛŋ ɡaɪz/ |
| good guy | /ɡʊd ɡaɪ/ | /ɡʊɡ ɡaɪ/ |
| can come | /kæn kʌm/ | /kæŋ kʌm/ |
| went quickly | /wɛnt ˈkwɪkli/ | /wɛŋ ˈkwɪkli/ |
Alveolar → palato-alveolar (mostly via /j/-coalescence)
Final /s/, /z/ before /j/ shift to /ʃ/, /ʒ/ — this is technically coalescence (two segments fusing) rather than pure place assimilation, and the same examples appear in the coalescence section below. We list them here because the surface effect (alveolar → palato-alveolar) is the same.
| Phrase | Citation | Assimilated |
|---|---|---|
| this year | /ðɪs jɪr/ | /ðɪʃ jɪr/ — /s/ + /j/ coalesces |
| miss you | /mɪs ju/ | /mɪʃu/ |
| because you | /bɪˈkəz ju/ | /bɪˈkəʒu/ |
Production tip
Place assimilation is automatic for natives — your mouth wants to do it. The Russian L1 task is to stop preventing it through over-careful articulation. Let your articulators take the shortest path between sounds, and assimilation happens naturally.
3. Voice assimilation and the “hafta” family
When a voiced consonant meets a voiceless one, the voiced one often devoices. The most familiar case: have to → hafta.
Voice-devoicing examples
| Citation | Fast (devoiced) |
|---|---|
| have to | hafta /ˈhæftə/ — /v/ → /f/ |
| has to | hasta /ˈhæstə/ — /z/ → /s/ |
| used to | useta /ˈjustə/ — /zd/ → /st/ |
| supposed to | supposeta /səˈpoʊstə/ — /zd/ → /st/ |
| good times | /ɡʊtˌtaɪmz/ — /d/ → /t/ before /t/ |
Voiced flap and voice mixing
The flap /ɾ/ is voiced regardless of underlying /t/ or /d/. So latter and ladder are phonetically /ˈlæɾɚ/ — homophones. Across word boundaries: get out of /ˈɡɛɾaʊɾəv/ has all-voiced flaps even though /t/ is underlyingly voiceless.
Russian L1 problem
Russian has strict voicing assimilation in the other direction (final devoicing: голод → /ɡolot/, луг → /luk/). Russian carries final devoicing into English: good → /ɡʊt/, had → /hæt/, love → /lʌf/. This is one of the most persistent Russian L1 markers at C1. The fix is to preserve voicing on final voiced consonants in English, even though Russian phonology fights against it.
4. Coalescence — /tj/, /dj/, /sj/, /zj/
When /t/, /d/, /s/, /z/ meet /j/ (typically in you, your, ya), they merge into single palato-alveolar consonants:
- /t/ + /j/ → /tʃ/
- /d/ + /j/ → /dʒ/
- /s/ + /j/ → /ʃ/
- /z/ + /j/ → /ʒ/
Coalescence examples
| Phrase | Citation | Coalesced |
|---|---|---|
| did you | /dɪd ju/ | /ˈdɪdʒə/ — “didja” |
| would you | /wʊd ju/ | /ˈwʊdʒə/ — “wouldja” |
| got you | /ɡɑt ju/ | /ˈɡɑtʃə/ — “gotcha” |
| meet you | /mit ju/ | /ˈmitʃə/ — “meetcha” |
| as you | /æz ju/ | /ˈæʒu/ (coalescence) → /ˈæʒə/ (further reduction when you destresses to schwa) |
| miss you | /mɪs ju/ | /ˈmɪʃə/ |
| can’t you | /kænt ju/ | /ˈkæntʃə/ |
| this year | /ðɪs jɪr/ | /ˈðɪʃɪr/ |
| how’s your | /haʊz jɔr/ | /ˈhaʊʒɚ/ |
Production tip
Coalescence is so automatic that resisting it sounds careful and unnatural. Let /t/ + /j/ become /tʃ/. Russian L1 speakers often pronounce did you as a careful /dɪd ju/, which sounds like reading aloud. The natural form is /ˈdɪdʒə/ — produce it that way.
5. Stacked reductions in real fast speech
In conversational AmE, multiple reductions stack on the same utterance.
Example: Did you eat yet?
| Stage | Form |
|---|---|
| Citation | /dɪd ju it jɛt/ |
| Coalescence | /dɪdʒu it jɛt/ |
| Reduction | /dʒu it jɛt/ — initial /d/ elided in /dʒ/ |
| Further coalescence with eat | /dʒu wit jɛt/ |
| Final | /dʒit‿ʃɛt/ — “j’eat yet?” |
A four-word question becomes three syllables. C1 learners must train themselves to decode this without slowing down the speaker. Production at this level of reduction is optional — comprehension is mandatory.
Example: What are you doing?
| Stage | Form |
|---|---|
| Citation | /wʌt ɑr ju ˈduɪŋ/ |
| Reduced | /wʌtʃə ˈduɪn/ — “whatcha doin‘“ |
| Further | /wʌtʃə ˈduwn/ |
Example: I’m going to give you some.
| Stage | Form |
|---|---|
| Citation | /aɪm ˈɡoʊɪŋ tu ɡɪv ju səm/ |
| Reduced | /aɪmənə ˈɡɪvjə səm/ — “I’mma giveya some” |
| Further | /aɪmə ˈɡɪvjə səm/ |
AmE-specific reduction patterns
- AmE flaps everything intervocalic; BrE preserves /t/ /d/.
- AmE drops /t/ after /n/ in casual speech: winter /ˈwɪnɚ/, internet /ˈɪnɚnɛt/, twenty /ˈtwɛni/.
- AmE has more aggressive function-word reduction than BrE: gonna, wanna, gotta, oughta, hafta, useta, sposta, kinda, sorta.
- AmE drops /ð/ in some contexts: this is /sɪz/ in very fast speech.
Common L1 Russian interference
- Final devoicing carried from Russian: good /ɡʊt/, had /hæt/, love /lʌf/. Single most persistent C1 marker.
- Over-articulation of /h/ in he, him, her, have. Native: drop them.
- No coalescence — did you as /dɪd ju/ instead of /dɪdʒə/.
- No place assimilation — good boy with crisp /d/ instead of /b/.
- No /t/ deletion in clusters — last week with audible /t/.
- Over-careful function words — to, of, and, for with full vowels instead of schwa.
Listening strategy
Find an AmE podcast at conversational speed (not news anchor speed). Transcribe a 30-second segment phonetically, marking every reduction. You should see 20-40 reduction events per minute. Then play the segment at 0.75× speed and verify your transcription. Many learners are shocked at how much reduction is actually happening — much of what they thought they couldn’t hear was deleted or assimilated, not just spoken fast.
Common Russian-speaker mistakes
- Final devoicing on /d/, /b/, /v/, /z/, /g/. Wrong: good /ɡʊt/, love /lʌf/, had /hæt/. Right: preserve voicing — /ɡʊd/, /lʌv/, /hæd/. Why: Russian devoices final voiced obstruents; English does not, and this is one of the strongest Russian L1 markers.
- No /h/ deletion in function words. Wrong: full /h/ in I told her /aɪ toʊld hɝ/. Right: /aɪ toʊl‿dɝ/ with /h/ dropped. Why: AmE drops /h/ in unstressed he, him, her, have, has, had; preserving /h/ sounds careful.
- No coalescence /t/+/j/, /d/+/j/. Wrong: did you as /dɪd ju/, got you as /ɡɑt ju/. Right: /dɪdʒə/, /ɡɑtʃə/. Why: AmE coalesces /tj/ → /tʃ/ and /dj/ → /dʒ/ across word boundaries; absence sounds like reading aloud.
- No place assimilation. Wrong: good boy /ɡʊd bɔɪ/ with crisp /d/, in May /ɪn meɪ/ with crisp /n/. Right: /ɡʊb bɔɪ/, /ɪm meɪ/. Why: alveolar consonants assimilate to following bilabial or velar place automatically; preventing this sounds careful.
- No /t/ /d/ deletion in clusters. Wrong: last week /læst wik/, send me /sɛnd mi/. Right: /læs wik/, /sɛm mi/. Why: AmE deletes /t/ /d/ at the end of clusters before consonants.
- No “hafta/wanna/gonna” reductions. Wrong: I have to go /aɪ hæv tu ɡoʊ/. Right: /aɪ hæftə ɡoʊ/. Why: voice assimilation devoices /v/ to /f/ before voiceless /t/; to reduces to schwa.
- Citation-form pronunciation in casual contexts. Wrong: full citation pronunciation in casual speech, sounds like reading. Right: apply reductions appropriate to register — full forms for formal, reduced forms for casual. Why: native speakers shift register acoustically; over-careful articulation in casual contexts sounds artificial.
Summary
- AmE fast speech systematically reduces sounds via elision, assimilation, voice change, and coalescence.
- /t/, /d/, /h/ delete in specific phonological contexts; alveolar consonants assimilate to following place; voiced consonants devoice before voiceless ones; /t/+/j/ and /d/+/j/ coalesce to /tʃ/ and /dʒ/.
- Multiple reductions stack: did you eat yet → /dʒit‿ʃɛt/.
- Final devoicing carried from Russian is the single most persistent L1 marker — work to preserve voicing on final /d/, /b/, /v/, /z/, /g/.
- Comprehension at C1 requires automatic decoding of stacked reductions; production at C1 requires at least the common reductions (hafta, gonna, didja, wouldja, gotcha).
Next lesson: academic vocabulary pronunciation — stress patterns on AWL words across morphological families (analyze, analysis, analyst, analytical, analytically), Greek-Latin compounds, and suffix-stress rules.