Phonemic analysis is/was motivated by several considerations:
1. A need to overcome the massive detail that ever narrower
2. A practical need to create alphabetic systems appropriate for a given language.
3. A belief that people have a psychologically real notion of the sounds of their language, in a fashion that corresponds to phonemes, not to phones.
4. A recognition that there are constancies across dialects that involve objects more abstract than sounds; these objects are phonemes. (This is closely related to 2, and perhaps 3 as well).
A phonemic analysis is a process that takes as its input either (1) a set of utterances, transcribed phonetically or (2) a speaker of a language, and produces a set of symbols which represent distinct phonemes. (There are 7 further conditions: )
1. The set must be minimal, in the sense that there may be no smaller set of phonemes that satisfies the conditions for phonemic analysis.
2. It must be possible to represent every utterance of the language as a string of phonemes.
3. It must be possible to represent every utterance of the language as a string of phones, where phones are the symbols used in the phonetic transcription.
4. There must be a one-to-one relationship between the phonemes in (2) and the phones in (1).
5. It must be possible to establish rules of allophony: these rules specify the correspondences between the two levels of representation mentioned in (4). In particular, a rule of allophony says, "Phoneme M is realized as Phone P in context C", where context C says what sounds are present to the left or right (or both) of Phone P. More graphically:
M1 M2 M3
P1 P2 P3
Typical rule of allophony:
"M2 is realized as P2 when followed by P3, otherwise it is realized as P9"
6. Uniqueness: A phonemic analysis must provide a unique phonemic representation for any given phonetic representation.
Typographical convention: phonemes are placed inside slashes /phonemes/, and phones at the phonetic level are placed within square brackets: [phones].
I-Raising: /ay/ is realized as [^y] when followed by a voiceless obstruent (ptksf) in the same word, otherwise it is realized as [a:y].
A rule of allophony always ends with the statement, "otherwise it is realized as ..." – if only because that is the simplest way to state it (i.e., one realization can be stated without specifying the context).
What does a symbol mean?
Philosophy #1: A phonetic symbol is understood to represent a specific linguistic sound. A phonemic symbol has meaning only insofar as it relates to specific phones. The particular symbol used has essentially no other significance. The mean lies in the correspondence rules (rules of allophony).
Philosophy #2: Human beings are extraordinarily good at discovering the phonemic relationships (allophony rules) that structure language, and the realization that is given in the "otherwise" formulation of the allophony rule is the (psychologically) real sound-image that comprises the phoneme.
Ø Phonemic analysis is primarily practical: it is meant to be practiced.
In practice, students are taught to look at all pairs of sounds that are similar and to test whether these two sounds are in an allophonic relationship (i.e., whether they might be realizations of the same phoneme). This requires one to learn what it means for two sounds to be similar (of course, one could simply consider all pairs of sounds....). Next, one (successively, iteratively) looks at each pair of sounds, and asks whether that pair might be realizations of the same phoneme.
The best test of whether two phones might be realizations of the same phoneme is the minimal pair test: if we can find two words that are different words and which are identical, except that one contains phone P, and the other contains phone Q in the same position, then the two constitute a minimal pair with regard to the pair P and Q, and P and Q cannot be allophones of the same phoneme. E.g. I and e in English: pit and pet are different words; hence these two sounds cannot be allophones of the same phoneme. By contrast, the two sounds occur in French Canadian, where no such minimal pairs exist.
If we cannot find any minimal pairs distinguishing two phones P and Q, then we can look for a principle that will specify a context in which one of them is used (with the other used "elsewhere"). If we can find such a principle, then we have established that the two are allophones of a single phoneme. (Convince yourself that this is techically always possible if the corpus is finite.)
7. There is another possibility that phonemic theory allows which we have not discussed: two phones P and Q which are realizations of the same phoneme M may be in free variation (in some context, or in all contexts): a word containing the sound P may be changed by replacing P by Q, and the result is another acceptable pronunciation of the same word.
Notice that both the statement of free variation and the definition of minimal pair requires being able to know whether two words are the same or different.
Bottom-up philosophy: phonemics is deeply rooted in a conception of analysis beginning with the phonetic, followed by analysis at the phonemic, followed by analysis of morphemes, and so on. This bottom-up philosophy can be interpreted as scientific methodology or as psychological theory about human speakers – or both. If it is taken as a theory about language users, it is a theory of language hearers rather than speakers. (Why?)
I said above that with a finite corpus and no minimal pairs for sounds P and Q, it is always possible in principle to establish a phoneme that is realized as P and Q. But it is generally understood that the principles of allophony must have some phonological simplicity or naturalness to them.
This was generally understood to mean, in addition, that the phones P and Q must not be too different phonetically, but that phrase was never successfully defined. Phonemicists all knew that some such clause was necessary to prevent an analysis in which [h] and the velar nasal are allophones of the same phoneme. (Why?)
Bottom-up, hearer-oriented analysis
Hearers have access to the string of phones representing an utterance. Do they have access to the presence of word-boundaries separating these words? Is this an empirical question? – that is, is it possible that some languages provide phonetic cues to word-boundaries, and others do not?
This is important, because if allophony does not have access to word-boundaries, what do we do with the phonetic realization of high time? Which realization of the vowel of high do we find? Why not the other?
Generative phonology: speaker-oriented analysis
The speaker knows where the word-boundaries are, and what phonemes are being uttered, so let the conditioning on phoneme-realization be dependent on what is known to the speaker, not just on what the hearer will be able to hear. A cognitive analysis of what an adult, competent speaker does is not modeled by a linguist's discovery procedures (though a child's acquisition may be).
Ø What does I-Raising do before a flap [D]? Consider rider and writer
Generative analysis says that the conditioning environment is found at the "underlying" level (here, equivalent to phonemic level). Observe that in many, many cases, the conditioning could be stated at either the phonemic or the phonetic level. We have to look harder for cases where it matters.
What accounts for the relationship between [t] and [D]? Is it a rule of allophony? What accounts for the relationship between [d] and [D]? Is it a rule of allophony? Is the Uniqueness condition met? la[D]er
Consequence: a phonemicist accepting the conditions stated above cannot accept Flapping as a rule of allophony. A generative phonologist can accept Flapping as a phonological rule.
What is the relationship between I-Raising and Flapping? Generative phonologists assume that rules apply sequentially. Which must come first?
It is not logically necessary to assume that rules apply sequentially. Develop an alternative.