For speech-language pathologists

How do I take a conversational speech sample for articulation and carryover?

Published · Updated

The short answer

Record 5 to 10 minutes of natural talk on topics the child cares about, ideally ones that draw out the target sound, then transcribe it and score every attempt at the target, or every consonant for PCC. Conversation matters because single-word tests and conversation can give different answers: in one study of 61 children, they led to the same pass-fail decision on only 71% of consonants on average. Sample at baseline and periodically, not every session.

Why conversation matters for carryover

Carryover means using a sound at increasingly challenging levels, from syllables and words to phrases, sentences and conversation (ASHA). Word-level data can’t show whether that last step has happened.

There’s a cost. In a study of 20 children aged 3;0 to 10;5, conversational samples took about 3 times as long to elicit and transcribe as a tailored single-word task, which gave sufficient information for phonological evaluation. The authors still recommend a brief conversational sample as a check on how representative the word sample is (Masterson et al., 2005).

How to elicit a sample

ASHA lists storytelling or retelling, describing pictures, and conversation about a topic of interest as ways to elicit connected speech. For young, reluctant or less intelligible children, it suggests recording spontaneous conversation at home, and repeating the child’s utterances when possible to make later transcription easier (ASHA).

Scoring the target sound in connected speech

For carryover, the usual measure is percent correct for the target: correct attempts divided by all attempts at the target, times 100.

The session sheet in the free speech therapy data sheets lets you mark each attempt as you listen back.

Percent consonants correct from a connected sample

PCC is the percentage of consonants produced correctly in a continuous speech sample, introduced as a severity metric with four levels from mild to severe (Shriberg & Kwiatkowski, 1982). ASHA describes the steps: collect and phonetically transcribe a speech sample, then count the total consonants and the correct consonants (ASHA). Progress monitoring in speech sound therapy covers where PCC fits alongside probes.

How PCC and two variants score errors (Shriberg et al., 1997)
MetricDistortionsDeletions and substitutions
PCCAll count as errorsErrors
PCC-Adjusted (PCC-A)Common clinical distortions count as correct; uncommon ones as errorsErrors
PCC-Revised (PCC-R)All count as correctErrors

The 1997 authors recommend PCC when all speakers are 3 to 6 years old with speech delay, and PCC-R for comparing speakers of different ages and speech status; PCC-R has lower standard errors because distortions are hard to transcribe reliably. They also note that a speaker with only distortion errors can score as low on PCC as a child with mild speech delay (Shriberg et al., 1997).

For a child whose remaining errors are distortions, such as a lateral lisp, PCC-R won’t change as the sound improves, so track target-sound accuracy instead.

Recording and privacy

Scoring connected speech live is hard, so most samples are recorded. Treat the recording as part of the child’s record.

How often to sample

There’s no research standard for how often to take a conversational sample. Two practical anchors:

Before discharge, a sample with a different listener or setting, such as a parent at home, is a stricter test of carryover than one in the therapy room. See finishing speech therapy.

Questions SLPs ask

How long should a conversational speech sample be?

PCC was designed for a 5- to 10-minute conversational sample. For a single target sound, the number of attempts matters more: with only 20, each attempt moves the score by 5 percentage points.

Is a single-word test enough to measure carryover?

Not on its own. In one study of 61 children, tests and conversational samples gave the same pass-fail decision on only about 71% of consonants on average, and some errors appear only across word boundaries.

Should distortions count as errors in PCC?

In the original PCC, yes. PCC-Revised scores all distortions as correct, which makes it more reliable across ages but blind to distortion-only errors such as a lateral lisp.

How often should I take a conversational sample?

There’s no research standard. Because samples take about three times as long as a word task to elicit and transcribe, they suit baseline, periodic checks during generalization, and discharge better than every session.

Sources

Join the Early Adopter Program