Re: [tied] Re: Retroflexes in Sanskrit

Phonemicity (inasmuch as any sort of reality can be ascribed to it) is a mental phenomenon, and the evidence for phonemic status is always indirect, since native speakers have no conscious access to their "analytic engine". A phoneme is a member an abstract system, contrasting with any other member at least potentially, but not necessarily. The minimal-pair test may fail occasionally, as in the case of English /h/, /Z/ = "zh" and /N/ = "ng": as far as I can see, none of them contrasts with either of the others, but no-one in his right mind would argue that they are allophones of the same phoneme. The existence of some phonemes is more secure than that of some others, which is why we use terms like "marginal phonemes", "quasi-allophonic distribution" or "emerging/disappearing contrasts". In other words, there is a scale of robustness for putative phonemes, rather than an infallible "yes or no" algorithm. The analysis of some items may vary from linguist to linguist (and presumably from one speaker's unconscious mind to another). I don't know how to answer a question like "is English /OI/ one phoneme or a tautosyllabic vowel cluster?" so as to satisfy all my colleagues -- opinions vary and will probably vary for ever, since competing analyses are possible, each with its own merits.

In Old English front [æ], back [a] and the short diphthong [æa] were in complementary distribution in strong masculines and neuter nouns, but the allophony rules that govern their occurrence could be violated (or rather replaced with a different pattern) in feminine nouns, adjectives and verbs, and there were quite a few potential minimal pairs or even trios like earn [æarn] 'eagle' : ærn [ærn] 'house' : arn [arn] 'ran'. It seems they were "in the process of" being phonemicised in the language, but the process was aborted in its mid course, and they fell together again in Middle English.

If there is a choice, I tend to side with the splitters rather than the lumpers as a matter of methodological preference, since a "splitting" phonemic representation contains more explicit information. It's easier to lump than to tell apart what has been lumped (especially if the proposed allophony rules are not very natural or straightforward), which is why lumping can always wait, as far as I'm concerned. I also prefer an analysis that is less abstract and more surface-true to one that uses primarily theoretical arguments and aims at maximum parsimony in terms of phoneme-counting but has to complicate the formal machinery in order to derive the right surface forms. For these reasons I prefer to regard Sanskrit /t./, /t.H/, /d./, /d.H/ and /n./ as phonemes, while admitting that their status is far less secure than that of their dental counterparts and alternative analyses are possible for some of them, at least in theory.

Piotr

----- Original Message -----
From: richardwordingham

To: cybalist@yahoogroups.com

Sent: Tuesday, August 06, 2002 12:00 PM

Subject: [tied] Re: Retroflexes in Sanskrit

... My suggestion was that Vedic did not have retroflex _phonemes_, from which I would have excluded /s./, but allophones which could be identified when they became phonemic in later speech. The clearest example of this phenomenon is the palatal nasal /n~/, which does not contrast with the guttural nasal /N/. The latter phonemic, but barely so, rather as in English dialects (idiolects?) that rhyme 'finger' and 'singer'. However, I see from that list that the suggestion is wrong. The past participle from the root {vah}, < PIE weg^H 'convey', has past participle u:d.Ha (and infinitive vod.Hum), courtesy of Bartholomae's Law and the loss of voiced sibilant allophones before voiced plosives. I should have remembered this form! This loss also yields ni:d.a. Minimal pairs are few and far between, but one has been dug up on the Indology list, using an inflected form, for /d.H/. There may not be one for /d./, but given the environment in which it occurs, it is hard to see why it should not be phonemic. (I refuse to believe that the English word <hang> is anything like /hah/.) [t.] is more problematic. One might dismiss it as the allophone of /t/ after /s./ and the pre-juncture allophone of /s'/, /s./ or /d./, though it is not nice to allocate a phone to different phonemes depending on the environment. What's the current view on such hocus pocus? Neater hocus pocus would be to declare a /d./ phoneme with phonetically conditioned voicing to account for [t.] and [d.], voiced between vowels and voiceless after sibilants, otherwise following the normal plosive voicing rules. [t.H] is like [t.], but simpler, as it does not occur prejuncturally. I trust it's not too unnatural to declare it an allophone of /d.H/. However, I think the distribution of [t.] invites the establishment of /t./ and /t.H/ as separate phonemes, available for use in new words, so my suggestion that Vedic did not have a retroflex order has little merit. The phonemicisation of the contrast of [n] and [n.] could have been much later. On the other hand, it could be simple hole-filling once the contrast in the plosives had been established. (I presume that would be firmly established by -nn- v. -rn.- becoming -nn- v -n.n.- in Prakrit.) It was interesting to see Dravidian blamed for the later *loss* of the contrast! (The reason given is that Dravidian has a dental v. alveolar v. retroflex contrast. Indic dentals sound alveolar to Dravidian speakers; Dravidian alveolars sound retroflex to Indic speakers. There are complications to this picture.) The other source of Prakrit retroflexes was as I suspected - /rt/ > /t.t./, along with /r.t/ > /at./. However, /s.t./ will also have been a rich source. Regards, Richard.

Your use of Yahoo! Groups is subject to the Yahoo! Terms of Service.