When the sounds existing in a given language have been analysed, it is always found that some of them may be regarded as variants of others, which they replace in certain well-defined conditions. A family consisting of one distinctive sound of the language together with the variants which represent it in particular sound-combinations is called a "phoneme." The nature of a phoneme is best shown by examples. The three k's in keep, cool, cot are different sounds, but for practical purposes they may be treated as a single entity; they are said to belong to the same phoneme. The precise k-sound used in any particular case is determined by the nature of the neighbouring sounds, and the different varieties are not used for the purpose of distinguish ing English words. Similarly the three t-sounds in eight, eighth, train belong to one phoneme, because the variants in eighth (phonetically eit0), train, are dependent upon the neighbouring sounds 0, r. The French sounds 1 and 1 (voiceless 1) are mem bers of the same phoneme. 1 is used in French only in final position when a consonant precedes, as when the word boucle occurs at the end of a sentence ; ordinary voiced 1 does not occur in this position, and consequently a Frenchman naturally regards the two very dissimilar sounds 1 and 1 as a single entity.
It often happens that sounds which belong to one phoneme in one language constitute separate phonemes in another language; they are consequently used for the purpose of distinguishing words in the second language though not in the first. Thus 1 and 1 are separate phonemes in Welsh, Burmese and Zulu, though they constitute only one phoneme in French (1 is transcribed 4 when it is a separate phoneme). Again h and c (the German ich-sound) are separate phonemes in German, but c is a member of the
in Japanese (the sound used when i follows). The French sounds of e and é are separate phonemes in French and are used for distinguishing words in that language ; but similar sounds occur in Russian and in Zulu as members of one phoneme, their use being determined in Russian by the nature of the adjacent consonants and in Zulu by a principle of vowel-harmony.
Phonemes are families of sounds occurring in a language as pronounced by a single individual. They must be distinguished from diaphones (see below) which contain sounds used by two or more different speakers of a language.
The simplest system of phonetic transcription of a language is that which provides one symbol for each phoneme of that language. Such a system is known as "broad" transcription. It is unambiguous to the native and to anyone who knows the rules governing the use of the subsidiary members of the phonemes.
More detailed transcriptions, providing symbols for subsidiary members of phonemes, are called "narrow" transcriptions. They are useful in comparative work, dialectology, etc.
Speakers of the same language often do not all use exactly the same sounds. Thus the vowel element of the word day is pronounced in several different ways according to the locality and to the social position of the speaker (phonetically de :,
des, dee, thei, etc.). In such cases it is convenient to have a term to denote the family of sounds consisting of a sound of a language as pronounced by an "average" speaker together with other sounds which are used in place of it by other speakers. Thus the various vowel sounds used in the word day by different speakers of English are said to constitute one diaphone in the English language.
In nearly every language the number of phonemes is greater than the number of letters in the Roman alphabet. It is therefore impossible to represent most languages unambiguously by means of the ordinary Roman letters only. So any romanic phonetic system, whether designed as a current orthography or as a transcription for the use of language learners, must contain new letters to supplement those of the present Roman alphabet. It can be shown that, on psychological, typo graphical and pedagogical grounds, specially designed letters are better than ordinary Roman letters with diacritical marks at tached. In the designing of new letters several factors have to be taken into consideration, the chief of which is that the printed forms of the letters should harmonize as well as possible with the ordinary Roman characters, so that the general appearance of the printed page may be satisfactory. It must also be possible to write the new letters with ease. It is further desirable that the letters should have italic forms which do not differ from the Roman forms in any essential part of their outline. To ensure legibility it is necessary that each letter should be as different as possible from every other letter. For typographical reasons the letters should, as far as possible, be designed so as to avoid the necessity of kerned types. Kerns at the top or bottom of the type are entirely inadmissible. Kerns at the side are un desirable and must be resorted to as sparingly as possible ; but it does not seem possible in a comprehensive alphabet to avoid them altogether. Lastly, it is a great advantage that the phonetic alphabet as a whole should be international, that is to say that each letter should always be used with the same or approxi mately the same value.