"Phonemic" distinctions induced by morphology

Question

As a native speaker of Bengali (Calcutta dialect), I found it quite surprising that this Wikipedia article on Bengali phonology asserts, "for most speakers /s/ and /ʃ/ are phonemically distinct." I had always assumed that only those dialects which have shifted /tɕ/ and /tɕʰ/ to [s] can be fairly analyzed to have such a phonemic distinction. The page gives examples such as সিরকা (sirakā <-- سرکه from Persian) /ˈsiɹkä/ (a word I've never used) and বাস্ (bās <-- bus, from English) /bäs/ which indeed are often (though not always or consistently) pronounced with [s] as distinct from /ʃ/ (phonetically [s̠], a laminal retracted sibilant, seldom palatalized or domed), but are obviously loanwords. What really caught my attention, however, is the pair আস্তে (āste: slowly, softly) /ˈäst̻e/ and আসতে (āsate: to come) /ˈäʃt̻e/. For me, they are homophones, both [ˈäs̻t̻e], but I do know people who consistently distinguish the two. And, I can also mention কাস্তে (kāste: reaping-hook, scythe) /ˈkäst̻e/ and কাশতে (kāśate: to cough) /ˈkäʃt̻e/, which even I sometimes distinguish. In fact, the chapter on Bengali in The Indo-Aryan Languages (Routledge (2003), eds: George Cardona and Dhanesh Jain) confirms this 'phonemic' distinction.

My point is that in both examples, the variation between /s/ and /ʃ/ can be explained by morphemic breaks. The default (laminal) post-alveolar sibilant /ʃ/ (or as mentioned before, [s̠]) assimilates to laminal denti-alveolar [s̻] adjacent to other laminal denti-alveolar consonants, such as /t̻/ and /t̻ʰ/ (and for some people, even [n̻]); this gives আস্তে (āste) with [s̻]. The word আসতে (āsate), however, is an inflection of the verb আসা (āsā: to come) /ˈäʃä/, with the inflectional ending -তে (-te) suffixed to the stem আস- (āsa-) /äʃ/, and therefore, assimilation is not triggered. The same analysis applies to কাস্তে (kāste) and কাশতে (kāśate), where the suffix -তে (-te) is added to the stem কাশ- (kāśa-) /käʃ/. If this is true, why are the authors suggesting a phonemic distinction? The distribution of the two sounds seems to be entirely predictable (even though minimal pairs clearly exist). Most monolingual Bengali speakers are, in fact, unable to distinguish /s/ and /ʃ/ from other languages, such as Hindi-Urdu and English.

My question, of course, is not Bengali specific, but the above simply serves as an example. My understanding was that (roughly speaking) two sounds are distinct phonemes if they are distinguished in similar environments, and no other factor can explain the difference. Even if other distinguishing features exist, does the presence of different sounds in minimal pairs deserve to be granted phoneme statuses, or is this a manner of preference on the part of the phonetician in charge?

Answer 1

The problem that you're encountering is due to different definitions of "phoneme", a problem which is about as old as the tern "phoneme" itself. Some questions here that center around the definition of phoneme are this, this, this, and most recently this. Formal views of the phoneme can be broadly classified as "data-oriented" versus "derivation-oriented". The former view starts with only narrow phonetic transcriptions and some indication "Utterance X is / is not the same word as utterance Y" (for example "heart" said one time is the same word as "heart" said a second time; "heart" is a different word from "hearts"). The data-oriented view then looks for patterns within the narrow transcriptions and extracts generalizations like "aspirated [tʰ] only appears before a vowel", "aspirated [tʰ] does not appear after [s]", "[t] appears after [s]" and so on, arriving at the conclusion that [t] and [tʰ] are in complementary distribution, and a rule can be given for stating when you get one versus the other (looking at the narrow transcriptions).

The derivation-oriented view is what you might call "holistic" in that it looks at narrow phonetic transcriptions, but also considers hypotheses about the phonological source of those transcriptions (underlying forms), also the syntactic and morphological status of an utterance. It especially shifts the focus from asking "is it the same word?" to asking "is it the same morpheme?", so while "heart" and "hearts" are distinct words, they contain a common morpheme. Accordingly, the status of being a "phoneme" in this kind of approach is not a statement about sound relations at the surface level, it is a statement about whether certain sounds occur in underlying representations.

Because of the nature of derivational mechanics, these two approaches to "phoneme" can give different classifications to sounds. They have in common the idea of "being able to predict", but differ in what the basis of the prediction is. The data-oriented perspective looks only at the surface form, and cannot tell whether [ʃt] comes from /sVt/ versus and [st] comes from /st/; or from /st/ versus /s+t/ – from that perspective, there is a phonemic contrast. If you expand the range of things that can be considered in judging "predictability" beyond that which can be directly heard as in the derivation-oriented view, you have greater "predictive power" and thus fewer things will be "phonemic". The confusion rests on the second part of your summary: "no other factor can explain the difference" is not universally accepted.

It is impossible to determine which definition is "correct", since there is no agreed-on independent means of determining whether two sounds "are distinct phonemes", or not.