¹ Y. R. Chao, Non-uniqueness of Phonemic Solutions of Phonetic Systems 367 (Bull, of the Inst of History and Philology, Academia Sinica, 4.363-98). I do not wish to imply that Chao’s definition is not justified; a definition must be broad to include the range of usage of this term, as was Chao’s intention.

² The criterion of totality is mentioned by Twaddell (On Various Phonemes), Lang. 12.56. As I conceive totality, it involves material (sounds), permutations (combinations of sounds), and pattern (phonetic and permutational relationships of sounds).

³ The material on which this paper is based was taken from my own experience and from that of colleagues, and in a few cases from printed sources. In practically all cases, I have had the opportunity of checking the data with native informants. Colleagues to whom I am indebted include Prof. Sapir, Prof. Lipari, M. B. Emeneau, M. R. Haas, G. Herzog, J. Kepke, G. L. Trager, C. F. Voegelin. Printed Sources which need to be mentioned specifically are: Trubetzkoy, Die Konsonantensysteme der ostkaukasischen Sprachen, Caucasica 8.1 ff.; Voegelin, Shawnee Phonemes, Lang. 11.23-37; Swadesh, The Phonetics of Chitimacha, Lang. 10.345-62.

Forms are quoted in phonemic transcription, one symbol for each phoneme. The orthography is thus in terms of my conclusions regarding the phonetic system of each language, but I try to state all the pertinent phonetic and distributional facts in each case, and I trust the reader’s judgment will not be influenced by my orthography. In the use of symbols, the aim was to follow general usage, but it was necessary to adapt the symbols to the requirements of the particular languages.

Since this paper was submitted there has appeared Trubetzkoy’s Die phonologischen Grundlagen der sogenannten ‘Quantität’ in den verschiedenen Sprachen, Scritti in onore di Alfredo Trombetti 155-74 (Milano, Ulrico Hoepli, 1936). In the main, the conclusions agree. My emphasis is on the relative nature of all pho- nemics (see my last paragraph). Trubetzkoy (170) says: ‘Bei den geminierten Konsonanten setzt die Zuordnung des Einsatzes and des Absatzes an verschiedene Silben (bezw. Moren) die gesonderte Existenz des Anfangs- und des Schlussteils voraus. Bei jenen “langen” Konsonanten aber, die dieser Bedingung nicht entsprechen, können Einsatz und Absatz nicht als verschiedene Punkte, sondern nur als ein Ganzes gewertet werden’. Such rules, it seems to me, cannot have general validity. There will be phonetic systems in which they are contradicted. I also cannot agree with Trubetzkoy’s conclusion (173) that quantity as such plays no part in phonemics. If ‘das Sprachgebilde ist ja zeitlos’, then it cannot have syllabic division or distinction between beginning and end of sounds. And if it has not the phonetic feature of duration, then must it not also lack every other phonetic feature, including intensity, which Trubetzkoy does recognize? That length does not ordinarily occur as the sole phonetic differentia, I recognize.

⁴ Raised ¹ indicates primary stress with falling pitch, ² indicates primary stress with rising pitch, indicates secondary stress. Stress mark after the vowel indicates that the vowel is long, stress mark after the first post-vocalic consonant indicates that the vowel is short and the consonant long.

⁵ Examples like the last three show that long vowels may be followed by more than one consonant, and contradict the rule given in Björkhagen, Modern Swedish Grammar 34: ‘In a stressed syllable the vowel is long if it is followed by one consonant. The vowel is short if it is followed by more than one consonant.’

⁶ Meinhof, Die Sprachen der Hamiter 87 ff. (Hamburg, L. Friederichsen 1912)., Meinhof does not tell how the geminates are pronounced.

⁷ According to Grandgent and Wilkens, Italian Grammar 6 (Boston, Heath, 1915), s in groups of s + consonant goes with the following syllable. That this traditional rule is incorrect was brought to my attention by Prof. A. Lipari.

⁸ G. L. Trager informs me that palatalized nj lj are always long in medial position. These long medials are then in complementary distribution with the short initial n_j l_j. The situation presents an interesting complication, since complementation is one of the characteristics of positional variants. If the rest of the system supported it, we could consider the long and short sounds as phonemically equivalent. As it is, consistency with the great preponderance of cases requires that long medial n_j l_j be considered geminate clusters: n_j l_j are defective, occurring only initially (in rather few words) and in medial geminate clusters.

⁹ Trubetzkoy reported this situation in liquids and nasals in some Caucasic languages, but these sounds are geminates rather than unit phonemes, as Trubetzkoy (op. cit. 171) now recognizes.

¹⁰ See S. Einarsson, Parallels to the Stops in Hittite, Lang. 8.177-182.

¹¹ A Dravidian language of southwestern India. The facts are given for common Malayalam. Elegant Malayalam, employing many Sanscrit and English words, has a considerably larger number of phonemes and allows a whole new set of consonant clusters. In consequence, the case of the strong consonants is more complicated, but they have the same status of independent unit phonemes.

¹² In medial position, they tend to be partly voiced and spirantal.