Linguistics Stack Exchange is a question and answer site for professional linguists and others with an interest in linguistic research and theory. It only takes a minute to sign up.
Sign up to join this community
Hypothetically, When describing the sound system of an unknown language, is there any scenario in which you may perform a phonetic analysis before a phonological one?
I guess the standard approach is phonological, using minimal pairs, and from there postulate a phonological system and only after make a phonetic analysis.
But are there exceptions in which a phonetic analysis proceeds a phonological one?
First, you need to explain what you mean by "phonetic", "phonological" and "analysis". You also need to explain what you mean by "before". To simplify matters, I reduce "phonetics" to acoustics, because that provides the necessary and sufficient basis for any statement about sound systems of spoken languages (and I also restrict the discussion to spoken languages).
In order to analyze a language at any level, you need data, which comes in the form of acoustic waveforms. The first piece of analysis that you perform is a segmentation, where you approximately divide the waveform into sub-pieces that correspond to "segments". Now you (Teusz) need to explain what you mean by "phonological" versus "phonetic". Presumably, you agree that anything that deals with numeric measurements of the waveform, such as the duration of a sub-part defined by certain properties, or anything having to do with frequencies or amplitude, these things are all "phonetic". Questions about rule-ordering, underlying forms, cyclicity and so on, are questions about phonology. It is completely unclear where to slot questions such as "is that a cluster [th] or an aspirated stop [tʰ]?"; "is that [d] or [t]?"; "is that sequence [tua] or [twa]?", along the phonetic-phonology divide.
The basis of all analysis is a phonetic object, the acoustic waveform, but the very first analytic product is a crude phonological categorization into segments. It is obviously premature to be thinking in terms of contrastiveness at this point: the initial goal is to develop an adequate system for reducing the massively variable acoustic waveform to vastly fewer discrete symbols. Is this a phonological analysis or a phonetic analysis? The input of analysis is a phonetic thing, but the output, a discrete segment symbol, is a phonological thing. So it depends on what you mean by "phonetic analysis" versus "phonological analysis".
Suppose the first word you hear is nivaka "knife". The objective fact is that there is about 30 msc VOT after the release of the stop, though you are not yet doing an acoustic analysis. There is no phonetic basis for deciding whether to record this as [nivakʰa] versus [nivaka]. It's not until you encounter ehalakha "basket" with 70 msc of VOT that you understand that nivaka and ehalakha contrast in terms of the phonetic property of VOT duration (a.k.a. aspiration). The thing that you have to attend to in the data is a phonetic property, namely VOT, and the product of your observations is a phonological fact, namely a contrast.
Discovering this contrast allows, indeed forces you to review your data to discover that you had previously encountered [kʰ] in solokho "mung bean", which you believed to be soloko. Since you happened to be recording everything, you determined that the speaker produced a somewhat atypical token with 50 msc VOT. There is no absolute minimum or maximum VOT value that defines phonetic aspiration, as shown by the fact that the aspirated stops of Thai have a shorter VOT than the unaspirated stops of Navaho (both languages have an aspiration contrast). One consequence of the fact that phonetic values of segments are not absolute is that in lieu of a contrast, you will have a hard time justifying a transcriptional choice. In very many languages, the highest back vowel sounds somewhere between Jones-defined [u] and Jones-defined [o]. IPA letters like [u], [ʊ], [o] represent ranges of formant values, and any attempt to numerically define the range presupposes a phonological categorization -- it would be invalid to include the values of the contrastive phoneme /ʊ/ from a language having the contrast /u, ʊ/ in a data-pool of formant values of [u]. It would be equally invalid to exclude languages which have no such contrast and where the highest back vowel sounds more like the Norwegian high back round vowel in the word "2".
The problem with the word "before" is that it may falsely lead you to believe that you must create a repository of phonetic conclusions which you can subsequently further analyze into phonological conclusions. There is no strict separation of levels of analysis: creation of some phonological conclusions is a prerequisite for creating phonetic conclusions, which can be used to justify further phonological conclusions. The basis of analysis in sound systems is a phonetic object, the acoustic waveform, but the kind of analytic object that results from doing the analysis may be phonetic or phonological, depending on the methods that you use.
Well, of course you have to start with phonetics, when you're just starting work with an informant. You can hardly start with phonemics, because there is no way for you to know what the phonemes are. You could be lucky and have an informant who is already literate in his own language, so he could simply tell you about the phonemes, but then you wouldn't be doing a phonemic analysis -- he would be telling you the analysis.
It is arguable that you can't do an adequate phonetic analysis, though, without also doing a phonemic analysis. In The Sound Pattern of English, Chomsky and Halle define what they mean by a phonetic representation by including a requirement that it be a transcription by an expert who knows the language. I suppose they have in mind the difficulty non-native speakers sometimes have in hearing some distinctions, like English stress (perhaps because some such distinctions are mistakenly thought to be phonetic).
It is also possible that in first confronting an unknown language, we proceed by at first assuming an incorrect phonemic analysis, and then gradually refining that, as we learn better how to fit the sounds we're hearing into our evolving understanding of the phoneme system.
Imagine a situation where you have a recording of people speaking in some language, but you have no possibility to talk to these people. You can't ask them questions that would lead to finding minimal pairs. In such a scenario, I don't see any option other than starting with the phonetic analysis: try to transcribe the speech signal phonetically, as precisely as you can, and only then look for patterns in the transcription, which may lead to hypotheses about the phonological system of the language.
