Phonetic transcription entails using special symbols to create a precise written record of an individual's speech. The symbols that are most commonly used are those of the International Phonetic Alphabet (IPA), first developed in the 1880s by European phoneticians. Their goal was to provide a different symbol for each unique sound, that is, to achieve a one-to-one correspondence between sound and symbol. For example, because [s] and [∫] are phonemically distinct in some languages, such as English, they are represented differently in the phonetic alphabet. Thus, the elongated s is used for the voiceless palatoalveolar fricative, as in [∫u]), to differentiate it from the voiceless alveolar fricative, as in [su].

The IPA has undergone several revisions since its inception but remains essentially unchanged. In the familiar consonant chart, symbols for pulmonic consonants are organized according to place of articulation, manner of articulation, and voicing. Nonpulmonic consonants, such as clicks and ejectives, are listed separately, as are vowels, which are shown in a typical vowel quadrangle. Symbols for suprasegmentals, such as length and tone, are also provided, as are numerous diacritics, such as [s] for a dentalized [s].

The most recent version of the complete IPA chart can be found in the Handbook of the International Phonetic Association (IPA, 1999) as well as in a number of phonetics books (e.g., Ladefoged, 2001; Small, 1999). Illustrations of the sounds of the IPA are available through various sources, such as Ladefoged (2001) and Wells and House (1995). In addition, training materials and phonetic fonts can be downloaded from the Internet. Some new computers now come equipped with “Unicode” phonetic symbols.

Although extensive, the IPA does not capture all of the variations that have been observed in children's speech. For this reason, some child/clinical phonologists have proposed additional symbols and diacritics (e.g., Bush et al., 1973; Edwards, 1986; Shriberg and Kent, 2003). The extended IPA (extIPA) was adopted by the International Clinical Phonetics and Linguistics Association (ICPLA) Executive Committee in 1994 to assist in and standardize the transcription of atypical speech (e.g., Duckworth et al., 1990). The extIPA includes symbols for sounds that do not occur in “natural” languages, such as labiodental and interdental plosives, as well as many diacritics, such as for denasalized and unaspirated sounds. It also includes symbols for transcribing connected speech (e.g., sequences of quiet speech, fast or slow speech), as well as ways to mark features such as silent articulation. Descriptions and examples can be found in Ball, Rahilly, and Tench (1996) and Powell (2001).

When transcribing child or disordered speech, it is sometimes impossible to identify the exact nature of a segment. In such cases, “cover symbols” may be used. These symbols consist of capital letters to represent major sound classes, modified with appropriate diacritics. Thus, an unidentifiable voiceless fricative can be transcribed with a capital F and a small under-ring for voicelessness (e.g., Stoel-Gammon, 2001).

Relatively little attention has been paid to the transcription of vowels in children's speech (see, however, Pollock and Berni, 2001). Even less attention has been paid to the transcription of suprasegmentals or prosodic features. Examples of relevant IPA and extIPA symbols appear in Powell (2001), and Snow (2001) illustrates special symbols for intonation.

Broad or “phonemic” transcriptions, which capture only the basic segments, are customarily written in slashes (virgules), as in /paı/or //. “Narrow” or “close” transcriptions, which often include diacritics, are written in square brackets. A narrow transcription more accurately represents actual pronunciation, whether correct or incorrect, as in [p^haı] for pie, with aspiration on the initial voiceless stop, or a young child's rendition of star as [t = aυ] or fish as [φıs].

How narrow a transcription needs to be in any given situation depends on factors such as the purpose of the transcription, the skill of the transcriber, and the amount of time available. As Powell (2001) points out, basic IPA symbols are sufficient for some clinical purposes, for example, if a client's consonant repertoire is a subset of the standard inventory. A broad transcription is generally adequate to capture error patterns that involve deletion, such as final consonant deletion or cluster reduction, as well as those that involve substitutions of one sound class for another, such as gliding of liquids or stopping of fricatives.

If no detail is included in a transcription, however, the analyst may miss potentially important aspects of the production. For instance, if a child fails to aspirate initial voiceless stops, the unaspirated stops should be transcribed with the appropriate (extIPA) diacritic (e.g., [p=], [t=], as in [p=i] for pea). Such stops can easily be mistaken for the corresponding voiced stops and erroneously transcribed as [b], [d], and so on. The clinician might then decide to work on initial voicing, using minimal pairs such as pea and bee. This could be frustrating for a child who is already making a subtle (but incorrect) contrast, for example, between [p=] and [b].

To give another example, a child who is deleting final consonants may retain some features of the deleted consonants as “marking” on the preceding vowel, for instance, vowel lengthening (if voiced obstruents are deleted) or nasalization (if nasal consonants are deleted). Unless the vowels are transcribed narrowly, the analyst may miss important distinctions, such as between [bi] (beet), [bi:] (bead), and [bĩ] (bean).

Stoel-Gammon (2001) suggests using diacritics only when they provide additional information, not when they represent adultlike use of sounds. For example, if a vowel is nasalized preceding a nasal consonant, the nasalization would not need to be transcribed. However, if a vowel is nasalized in the absence of a nasal consonant, as in the preceding example, or if inappropriate nasalization is observed, a narrow transcription is crucial.

Phonetic transcription became increasingly important for speech-language pathologists with the widespread acceptance of phonological assessment procedures in the 1980s and 1990s. Traditional articulation tests (e.g., Goldman and Fristoe, 1969) did not require much transcription. Errors were classified as substitutions, omissions, or distortions, and only the substitutions were transcribed. Therefore, no narrow transcription was involved.

In order to describe patterns in children's speech, it is necessary to transcribe their errors. Moreover, most phonological assessment procedures require whole word transcription (e.g., Hodson, 1980; Khan and Lewis, 1986), so that phonological processes involving more than one segment, such as assimilation (as in [g∧k] for truck), can be more easily discerned. (In fact, Shriberg and Kwiatkowski, 1980, use continuous speech samples, necessitating transcription of entire utterances.)

To facilitate whole word transcription, some clinical phonologists, such as Hodson (1980) and Louko and Edwards (2001), recommend writing out broad transcriptions of target words (e.g., /tr∧k/) ahead of time and modifying them “on line” for a tentative live transcription that can be verified or refined by reviewing a tape of the session. Although this makes the transcription process more efficient, it can also lead the transcriber to mishear sounds or to “hear” sounds that are not there (Oller and Eilers, 1975). Louko and Edwards (2001) provide suggestions for counteracting the negative effects of such expectation.

If a speech-language pathologist is going to expend the time and energy necessary to complete a phonological analysis that is maximally useful, the transcription on which it is based must be as accurate and reliable as possible. Ideally, the testing session should be audio-or video-recorded on high-quality tapes and using the best equipment available, and it should take place in a quiet environment, free of distractions (see Stoel-Gammon, 2001). Because some sounds are difficult to transcribe accurately from an audiotape (e.g., unreleased final stops), it is advisable to do some transcribing on-line.

One way to enhance the accuracy of a transcription is to transcribe with a partner or to find a colleague who is willing to provide input on difficult items. “Transcription by consensus” (Shriberg, Kwiatkowski, and Hoffman, 1984), although impractical in some settings, is an excellent way to derive a transcription and to sharpen one's skills. This involves two or more people transcribing a sample at the same time, working independently, then listening together to resolve disagreements.

Sometimes it is desirable to assess the reliability of a transcription. For intrajudge reliability, the transcriber relistens to a portion of the sample at some later time and compares the two transcriptions on a sound-by-sound basis, determining a percent of “point-to-point” agreement. The same procedure may be used for determining interjudge reliability, except that a second listener's judgments are compared with those of the first transcriber. Reliability rates for children's speech vary greatly, depending on factors such as the type of sample (connected speech or single words) and how narrow the transcription is, with reliability rates being higher for broad transcription (see Cucchiarini, 1996; Shriberg and Lof, 1991). Alternative methods of assessing transcription agreement may sometimes be appropriate. For instance, in assessing the phonetic inventories of young children, Stoel-Gammon (2001) suggests measuring agreement of features (place or manner) rather than identity of segments.

People who spend long hours transcribing children's speech often look forward to the day when accurate computer transcription will become a reality. Although computer programs may be developed to make transcription more objective and time-efficient, speech-language pathologists will continue to engage in the transcription process because of what can be learned through carefully listening to and trying to capture the subtleties of a person's speech. Therefore, phonetic transcription is likely to remain an essential skill for anyone engaged in assessing and remediating speech sound disorders.