There is nothing very new about the idea that speech sounds can be classified in terms of a limited number of features. Phonetic taxonomies have existed from the times of the earliest grammarians; and traditional consonant charts can be viewed as attempts to classify sounds in terms of features of place and manner of production. But with the development of the Prague school of phonology (Trubetzkoy, 1939) and the subsequent work of Jakobson and his collaborators (Jakobson, 1962; Jakobson, Fant and Halle, 1951; Jakobson and Halle, 1956) the discussion of the distinctive features of speech came to have more importance. As a result of the more recent work of Chomsky and Halle (1968), the role of distinctive features within phonological theory has become even more crucial. This paper will review the nature of phonological features in general, and will suggest a particular set of features which it might be appropriate to use in phonological descriptions of languages.

The most important tone sandhi is that of tone-3. Each of the two forms of bisyllabic sequences in Chinese, the quasi-iambic and the trochaic, has two types of tone sandhi pattern. As far as the quasi-iambic sequence is concerned, tone-3 followed by other tones (including the neutral tone) except another tone-3 changes to ½ tone-3. Tone-3 followed by another tone-3 changes to tone-2. As far as the trochaic sequence is concerned, tone-3 followed by the neutral tone derived originally from tone-3 changes either to ½ tone-3 as before any other neutral tone, or to tone-2 as before any stressed tone-3. Tone-3 followed by the first type of trochaic sequence changes to tone-2, and that followed by the second type changes to ½ tone-3. Tone sandhi patterns of the sequences formed by three stressed tone-3 syllables, though mostly determined by the IC, are rather inconsistent; they vary from speaker to speaker. There are two types of tone sandhi patterns of the sequences formed by more than three stressed tone-3 syllables. The first type is uttered with a momentary pause which has the function of separating one IC from another. The second type is uttered without the pause. Thus only the tone pattern of the final syllable remains unchanged. Whether a polysyllabic sequence is uttered with a momentary pause or not depends on the speaker. In general, due to the influence of stress distribution, 223-sequences and 2223-sequences may optionally change to 213-sequences and 2123-sequences respectively when they are spoken faster. This confirms that the description of sandhi patterns in the second paragraph of section 1 is correct.

The Daniel Jones Cardinal Vowel system, defined as it is by clear-cut and rigid rules, is part of the stock-in-trade of all IPA phoneticians. I wish to suggest that the present format of the system contains a contradiction so basic that any attempt to make use of it as a descriptive technique, in all but a very restricted number of cases, would be completely impossible. This contradiction, moreover, is so obvious that the fact that no phonetician has called attention to it until now is hard to explain, and the equally certain fact that descriptive phoneticians (including the present writer) regularly teach the use of the technique to students, and apply it themselves, is even more so.

In the second paragraph of Mrs. Waterson's article ‘Some views on speech perception’ (JIPA, 1, 2, 81), the reference ‘Fry, 1970’ should read ‘Fry, 1956’.

In the same number, in Dr. W. Jassem's article ‘Pitch and compass of the speaking voice’, the denominator of formula (4) on p. 65 should read s3, not s2.