A learner sees What are you going to do? and knows every word. In a Taipei café, the same sentence arrives closer to Whaddaya gonna do? and seems to contain two words rather than six. The vocabulary is already present in memory. Its ordinary spoken forms are missing. Listening lessons should teach a small set of frequent reductions as sound-to-meaning mappings, with full forms kept beside them.
This article is for an intermediate learner who understands graded recordings and loses familiar words in unscripted conversation.
Written words have spoken variants
Connected speech changes sounds at word boundaries. Speakers link consonants and vowels, weaken unstressed vowels, assimilate neighbouring sounds and sometimes delete a segment. Want to may contain a flap and a reduced vowel. The /h/ in tell him may disappear. These are patterned variants shaped by rate, prominence, register and surrounding sounds.
The aim is recognition across variants. A learner should connect careful going to, reduced gonna and the intended future meaning when each occurs. Treating one form as correct and the other as lazy speech gives poor guidance. Formal presentations often contain fewer extreme reductions than a quick exchange among friends, yet connected speech appears well beyond slang.
Spelling makes the mismatch easy to miss because the page preserves every word with clean spaces, while a speaker allocates time according to prominence and folds low-information syllables into their neighbours, leaving a listener to recover could you have sent it from a rhythm that never presents five dictionary forms in sequence. Print stays still. Speech compresses.
Choc Education's article on why listening breaks at word boundaries explains how known words vanish when listeners segment the stream incorrectly. Reduction makes those boundaries even quieter.
Taiwanese evidence locates the difficulty
Kuo tested 103 second-year English majors in Taiwan, grouped by their intermediate GEPT listening scores. They completed dictation on 24 recorded sentences containing consonant links, consonant-vowel links, /h/ deletion, contraction, palatalisation and flapping. The National Changhua University of Education study found that connected-speech modifications affected spoken-word recognition. Proficiency and modification type both predicted performance. Correct response rates for the high, middle and lower groups were .724, .576 and .434. One missed boundary can erase a whole clause before its meaning is evaluated.
The study was diagnostic and assigned no learners to a teaching treatment. Its English-major sample cannot represent every Taiwanese adult. The practical value lies in its error categories. A teacher can test six specific patterns instead of concluding that a learner simply needs more listening.
Older classroom evidence gives a cautious reason to teach the patterns. Brown and Hilferty assigned 16 learners to thirty 10-minute lessons on reduced forms and 16 to lessons on minimal pairs. Their RELC Journal experiment reported gains in listening for the reduced-form group, though the size of the benefit varied across three measures. Thirty-two participants is a small trial. It supports a trial lesson, not a universal claim about a complete course.
Start with six useful strings
Choose reductions from the learner's actual listening. A customer-service worker may meet these repeatedly:
| Careful form | Possible ordinary form | Meaning to retrieve |
|---|---|---|
| could you | /kədʒə/ | a request follows |
| did you | /dɪdʒə/ | a past question follows |
| have to | /hæftə/ | obligation |
| want to | /wɑnə/ or /wɒnə/ | intention |
| tell him | /telɪm/ | him receives the message |
| a lot of | /əlɑɾəv/ | quantity |
IPA can help a teacher and some learners. Audio must carry the lesson. Record each full form and one natural reduced version inside the same seven-word sentence. Preserve the surrounding sounds because a reduction learned in isolation may behave differently after a vowel or fricative.
I would rather teach six strings from tomorrow's client call than 28 reductions selected to fill a worksheet. Frequency and relevance decide the set.
Use comparison, concealment and retrieval
Begin with meaning. Play “Could you send it again?” once and ask the learner to choose between a request, an apology and a refusal. Replay it beside a careful version. The learner marks the part that changed, then sees the written sentence.
Hide the text.
Play four new sentences containing could you. The learner taps when the request frame begins and says the next likely action: check, send, confirm or hold. This keeps the reduction attached to a communicative prediction. Pure transcription can become a spelling contest after the sound has already been recognised.
Finish with production for perception. The learner says the careful version, then a comfortable connected version, and records both. Mouth movement draws attention to the lost /d/ and the new /dʒ/ sequence. Production remains optional for learners whose goal is listening. Accurate recognition has priority over copying a particular accent.
Keep variation inside the lesson
One recording can create a brittle memory. Use two speakers and two rates across eight examples. Include a careful form among the reduced ones so learners keep both routes. On the next day, change the verbs while preserving the frame: Could you check the invoice? Could you hold for a moment?
When the learner recognises a reduction from one speaker and misses it from another, keep the wording constant while changing voice, pace and surrounding sound, because the error now concerns generalising the sound pattern rather than remembering the sentence. Same words, new voice.
A 2021 Taiwanese study offers a useful warning about claims for video teaching. Chen and colleagues worked with 48 A2-B1 undergraduates for seven weeks and reported better connected-speech production on repeated and new reading passages. The open study also states that it had no control group and no delayed post-test. Improvement occurred, while its cause and durability remain uncertain. The treatment combined selected videos, teacher explanation, repetition, peer practice and feedback. “Use YouTube” is far too broad a lesson from those data.
Protect comprehensibility and choice
Reduced forms belong first in receptive practice. Learners can use careful speech and remain easy to understand. Pushing gonna or heavy linking into every utterance may sound mismatched in a board presentation, and forced reduction can damage consonants that listeners need.
Accent goals differ. Someone handling calls with American clients may want to produce a few common reductions. A researcher presenting at an international conference may choose clear full forms and train only to recognise variants from others. Both learners need access to the meaning when speech changes shape.
Tomorrow, take one 18-second clip from a real programme. Circle six written words the learner already knows, find the two that shrink in the recording, and build four new sentences around those same sounds.