Phones, phonemes and allophones

Starting at one of the main sets of concepts in phonetics and phonology, I bet most readers will roughly know these things. I will try and get into the physics of the entire thing a bit though.

phones are physical units of sound. They are what you actually produce while speaking. We also use this term to refer to transcriptions of phones, but already in the process of transcription is some of the reality lost - we're dealing with a map now, not with terrain. We surround them with square brackets.

phonemes are the classes to which we assign the speech sounds we hear, and possibly also the mental representation in some parts of our speech production of the sounds we intend to produce. Phonemes are surrounded by slashes.

allophones are usually written between square brackets, as they too are phones. Every allophone belongs, in a sense, to a phoneme. The classical definition of an allophone is a realization of a phoneme, or a phone that is heard/made to represent some phoneme. That is, when you are uttering the phoneme /p/, the sound you are making is a phone that belongs to /p/. Which particular allophone depends on the patterns these allophones appear by, e.g. in English /p/, when word-initial is [p^h].

In some languages, /p/ when between two vowels - in an intervocalic position - is rendered [b]. Similar alterations exist, such as some languages assigning [p^h]  and [p] to different phonemes. Thus, the phoneme is a language specific unit, and /p/ can mean quite different things when describing different languages - in English, /p/ includes [p] and [p^h], in some languages it includes [b] and [p] but not [p^h]. There is no such thing as a cross-linguistic phoneme - /p/ simply doesn't mean anything unless we provide some point of reference - a language, a language family, a geographical region where languages have influenced each other's phonemic systems.

To what precision the allophones are given in some grammar or description depend entirely on with what the author is concerned - is he concerned with describing how /p/ and /b/ differ in some language? He might want to include, say, careful annotation for the length of the time it takes for the voicing on the subsequent vowel to start (voice onset time). Maybe he is describing the types of r-like sounds in a language, and even counts - using spectrograms - the average number of trills and some kind of average deviation that happen during, say, /r/ vs. /r:/.

 There are some more modern theories of phonology where the phoneme has been abolished. However, the concept is still by and large useful for conlangers.

Some books on phonology will claim that no allophone can belong to two phonemes, others will have no problem with it. Personally, I find forbidding them a rather reasonable idea, and do sometimes advocate it even vociferously.

I do realize different definitions (and thus essentially somewhat different concepts) exist, and it may be more elegant to describe, say, the situation where some sounds under some conditions are merged with some other sounds as though the 'merged' form is a shared allophone - e.g. final devoicing, using overlapping allophones: Hund - [hunt], Hunt - [hunt] - are the final [t]s here allophones of different consonants? Keep in mind that in languages with this phenomenon, the voiced form reappears when an ending is applied: [hunt], but [hunde] in the case of Hund, and the final t would remain unvoiced for Hunt even though a vowel was suffixed.

This is of course up to the conceptual tools we use to describe the situation - we can use definitions of phonemes that permit it or we can use definitions that forbid it. Depending on which approach we take, the elegance of our description of a language may change - and the degree of elegance may differ in different ways depending on what particular thing about the language or Language in general we're describing. One description may work better for the psychology of first language acquisition, another may work better for describing the morphophonology, a third may work better for explaining quirks in the orthography.

I will probably explain closer sometimes later why I personally dislike the idea of shared allophones and explain in closer detail the problems they case. For now, though, let me just caution that if overlapping allophones exist in your conlang, there should be a good justification for the overlap. Having arbitrary allophones belong to more than one phoneme is more often a sign of an incomplete understanding of phonemes than it is of a genuinely overlap-like situation. By now I do feel mandated to write a more careful post on this at some time.