Consonant-to-Consonant Linking in English: Shared Sounds, Held Sounds, and Word Boundaries

Learn consonant-to-consonant linking in English: when to hold a shared sound, move into the next consonant, and avoid extra vowels at word boundaries.

Sofia Laurent

FunFluen editor · Idioms and culture

Aug 30, 2026 11 min

The short answer

When two consonants meet across words, do not automatically make two separate launches or delete one: listen for whether the model holds a shared consonant, moves directly into a different consonant, or makes a real audible boundary.

You can say keep. You can say calm. Put them together and your mouth may quietly install a vowel between them: keep-uh-calm. Or you attack both consonants so hard the word space sounds like a traffic accident. The consonants are not the problem. The join is.

Start here: Hold, Move, or Boundary?

This is a practical learner framework, not an official phonetic taxonomy.

Three useful decisions at a consonant-to-consonant word boundary
Decision	What you are listening for	Useful mouth cue
HOLD	The same consonant occurs on both sides of the boundary and is sustained rather than restarted as two dramatic events.	Keep the articulators in position through the boundary; let duration carry part of the join.
MOVE	Two different consonants meet, or a stop moves straight into the following consonant without a separate obvious burst.	Form the first consonant, then move immediately into the second. Do not park a helper vowel between them.
BOUNDARY	The speaker actually makes the edge clearer through a pause, careful style, emphasis, correction, or phrasing.	Keep the real boundary. Connected speech is not a competition to remove every pause.

One warning before anything else: no separately audible release does not mean the first consonant was deleted.

HOLD: when two written consonants can share one long articulation

Identical consonants across a word boundary are where spelling most strongly tempts you to stop and restart. Often, that is unnecessary.

Baruch College’s consonant-to-consonant linking guide gives a very practical distinction. With repeated stop consonants, such as the /g/ at the boundary of big game, its teaching model is to hold the closure longer and release into the second word. With repeated continuous consonants, such as the /m/ across same message, the sound can be sustained rather than broken and relaunched.

University College London phoneticist John C. Wells describes the same broad behavior in more technical terms. For repeated plosives across a boundary, he describes a long hold with no audible release of the first plosive and no fresh audible approach to the second. Repeated fricatives, nasals, and liquids are described as being like single articulations that last longer. See Wells’ “Concatenation, coarticulation” discussion.

Try the contrast without timing yourself

Do not chase a magic number of milliseconds. The research does not justify one universal learner target.

big game: feel one /g/ closure stretched across the word edge, then release into game.
same message: keep the lips together and the /m/ voicing going through the boundary rather than stopping the sound, opening, and launching another /m/.
rough fight: in the Wells teaching description, the repeated /f/ can continue across the boundary rather than becoming two isolated hisses.
ten names: the repeated /n/ can likewise be sustained across the word edge.

There is evidence that duration really can matter here, but it is not a universal stopwatch rule. In a Journal of Phonetics study by Grace Oh and Melissa Redford, identical consonants spanning word boundaries were longer than matched singleton consonants in absolute duration. The experiment involved eight American-English-speaking University of Oregon undergraduates and focused on specific nasal materials, so do not turn that result into “every double consonant must be exactly X times longer.”

The useful learner target is simpler: do not accidentally shorten a meaningful shared boundary into the wrong phrase, and do not manufacture two full restarts when the model gives you one sustained articulation.

The missing burst is not necessarily a missing consonant

Stops such as /p, t, k, b, d, g/ involve a closure. In careful dictionary-style thinking, learners often expect the consonant to finish with a neat audible release. Real connected speech is less obliging.

At a repeated-stop boundary, the first closure may have no separately audible release because the articulation continues into the second stop. Baruch’s teaching material also uses a similar cue when a stop is followed by a different consonant: form the first stop, avoid releasing it as a separate event, and move directly into the following sound.

That distinction matters for listening. Suppose you hear a model of good friend in which the /d/ is formed but you do not hear a tidy little /d/ burst before /f/. Your ear may shout, “They deleted the D!” Not so fast. A consonant can be articulated without handing you a separate acoustic fireworks display.

No separate burst does not mean the consonant has entered witness protection.

So when listening, ask a better question:

Do I have evidence that the consonant truly disappeared?
Or do I simply lack a separate release event because the articulations overlap or move directly into each other?

This article does not try to own every case of elision or assimilation. If a sound genuinely changes or disappears in a particular recording, that is a different connected-speech diagnosis. Here, the point is narrower: do not label a stop “deleted” merely because you did not hear its release separately.

MOVE: different consonants, no waiting room in the word space

Now take a boundary where the consonants are different: keep calm, rich food, bad thing.

Wells’ UCL discussion gives these as examples of phrases where a final consonant needs to meet a following consonant without an inserted vowel. His paper was especially concerned with Japanese learners and is written from a British-English perspective, so it should not be treated as evidence that every learner has this difficulty. But the physical practice principle is broadly useful: if your own speech inserts a helper vowel at the word edge, practise the consonant-to-consonant transition itself.

The boundary ladder

Use one phrase. Do not repeat the whole sentence twenty times and hope the middle somehow fixes itself.

Find C1 and C2. In keep calm, the target edge is /p/ → /k/.
Form C1. Close the lips for /p/. Do not add a vowel after it.
Move to C2. Shift directly into the tongue position for /k/.
Rebuild the words. Say keep calm slowly while preserving that direct movement.
Put it back into a sentence. For example, use the clearly constructed practice line: Try to keep calm for a moment.
Return toward your normal pace. Speed is not the skill. Keeping the transition while the rest of the sentence returns is the skill.

If an “uh” appears at the boundary, shrink the task back to /p/ → /k/. The word space is a spelling convention, not a request for a new syllable.

BOUNDARY: sometimes the speaker really does pull the words apart

Now the useful correction to all that linking advice: do not force every consonant boundary to sound maximally compressed.

Oh and Redford’s study found that careful speech could make word-boundary identical consonants more explicit, including with pauses and pitch changes. That was a small, specific American-English experiment, not a map of every English variety, but it is enough to kill the simplistic rule “natural speech always merges the boundary.”

Imagine two clearly labelled situations:

Neutral connected production: somebody says big game as part of an ordinary sentence, with a sustained /g/ boundary.
Deliberate correction: somebody was misheard and now says BIG | GAME with a clearer break or stronger prominence.

The second version is not “bad linking.” It is doing a different communicative job.

This is also why you should copy the speaker you actually have. Accent, rate, emphasis, phrase structure, and speaking style can all affect how obvious a word boundary becomes. Recognition should be broad; production should be selective.

One exception worth knowing: do not force the shared-hold shortcut onto affricates

The phrase each chair looks like a perfect candidate for “same consonant on both sides, so just hold one longer.” Wells explicitly gives repeated affricates as an exception to that shortcut in his English description: phrases such as each chair and orange juice are described with two complete affricates.

You do not need a full lesson on affricates here. Just keep this protection against overgeneralizing:

Same-looking boundary does not guarantee same articulatory solution.

Did the consonant disappear? Diagnose before you imitate

Choose an answer before opening each explanation. There is no score; the useful result is knowing what evidence changes your next listen.

In a teaching model of “big game,” the /g/ closure is held longer across the word edge and there is one release into “game.” HOLD, MOVE, or deletion?

HOLD. Both words still contribute /g/. The useful cue is the longer shared closure, not two separate explosive releases.

In “same message,” the /m/ continues through the boundary instead of stopping and restarting. What is the useful diagnosis?

HOLD. This is the continuous-consonant version: sustain the /m/ through the edge. If you make it too short, you can weaken the acoustic cue that separates same message from a sequence such as say message.

You hear “good friend.” The first /d/ has no neat separate burst, but the speaker formed the stop before moving into /f/. Did /d/ disappear?

Do not call it deletion. This is a useful MOVE analysis: the first stop is articulated and the transition goes directly into the next consonant even though the release is not separately obvious.

You produce “keep-uh-calm” while trying to say neutral “keep calm.” What needs repair?

MOVE. The helper vowel created an extra syllabic event that was not part of the intended phrase. Practise /p/ → /k/ directly, then rebuild the words.

A speaker deliberately says “BIG | GAME” after somebody mishears them. Must you remove the pause because consonants should link?

No: BOUNDARY. A deliberate correction or emphasis can make the edge clearer. Connected speech is context-sensitive, and communicative intent outranks a practice slogan.

You see “each chair” and want to use the same long-hold shortcut as “big game.” Good idea?

Not in the Wells teaching description used here. Repeated affricates are treated differently: use two complete affricates rather than assuming the simple shared-hold pattern.

Do one boundary all the way into a sentence

Choose one real line with a consonant-to-consonant word edge. Then use this checklist:

I listened once before deciding what the boundary “should” do.

I wrote the two boundary consonants, such as /g/ → /g/ or /p/ → /k/.

I classified the model as HOLD, MOVE, or a real BOUNDARY.

I practised the transition without inserting a helper vowel.

I rebuilt the two words and then the full sentence.

I listened again and changed my classification if the actual speaker contradicted my prediction.

The last box is the important one. The speaker wins. Your diagram does not get an appeal.

Two learner repairs that show why the boundary matters

Common consonant-boundary mistakes, listener effects, and repairs
Learner production	Classification	What a listener may hear	Likely learner intent	Natural repair	When the original can be valid
keep-uh-calm, with a helper vowel inserted between the intended /p/ and /k/ boundary	Wrong for the intended neutral phrase when the added vowel is accidental	An extra syllabic event; the phrase may sound accented, hesitant, or momentarily harder to parse	Say keep calm clearly without losing either word	Form /p/, move directly to /k/, then rebuild keep calm with no helper vowel	A real hesitation or filled pause can occur in speech, but that is a communicative pause, not the target consonant transition
same message with the boundary /m/ shortened so far that the first word can approach the acoustic shape of say before message	Context-dependent, but a poor match when the learner intends the modelled shared /m/ boundary	The word boundary cue may weaken; in an unfavorable context the listener may need more information to recover same	Produce same message as the two intended words	Sustain the /m/ through the word edge instead of deleting the first word’s final consonant or fully restarting the second	Duration varies by speaker and style; the repair is not a fixed timing prescription

Notice that neither repair says “blend harder.” You are preserving the words while removing an unnecessary reset — or restoring a cue you accidentally erased.

Take the boundary from a diagram into real speech

Once the manual move works, the next problem is boring but real: replaying the exact same line enough times to hear whether a consonant is held, separately released, or carried straight into the next sound.

FunFluen can reduce that navigation friction around a supported subtitled line with sentence navigation, repeat, listen-first practice, and a later speaking pass. If a boundary is genuinely hard to inspect, a slight temporary speed reduction can help before you return toward the original pace. Use that as deliberate practice support, not as a promise of perfect accent scoring.

Take one consonant boundary from listening into speaking. The first step after the click is choosing a general English speaking-practice path; this exact consonant-boundary lesson is not promised to be preloaded, and speaking practice depends on suitable original audio and subtitles.

The space is spelling. The boundary is speech.

Consonant-to-consonant linking in English is not one instruction to “blend.” It is a set of boundary decisions.

HOLD when the model sustains a shared consonant. MOVE when one consonant transitions directly into another without a helper vowel or unnecessary restart. Keep a real BOUNDARY when careful speech, emphasis, correction, or phrasing makes it audible.

And keep one distinction taped mentally above the whole exercise: no audible release is not the same thing as deletion.

That habit becomes much more useful when you are learning pronunciation from real media, because real speakers will not obey one diagram in every context. Listen first. Classify the boundary. Practise the physical move. Then let the speaker — not the written space — decide what survives.

No waiting room required.