ÍNDICE CAPÍTULO 7
7. PRESENTACIÓN Y ANÁLISIS DE LOS EPISODIOS DE LLUVIA ESTUDIADOS
7.1. PRESENTACIÓN DE RESULTADOS 1. Resumen de episodios de lluvia analizados
7.1.2. Caudales y volúmenes en episodios de lluvia
Recall that in an OT model the language input is subject to GEN – the Genera-tor which generates possible candidate representations, possibly innite, ac-cording to universal structural principles. The assumed universal structural principles in the case of language pertain to universal segment types, pho-nation types, universal syllabication and higher order structures, etc. For language, owing to a long history of investigation and research, one can as-sume quite a lot about possible representations. For instance, certain segment types are universally ruled out, e.g. labiodental stops; all languages must parse segments into syllables, universally, the nucleus of the syllables must be the most sonorous segment in the syllable, all syllables must be gathered into higher order structure which may be left or right headed (whatever that might mean to linguists) and the structure motivated by phonological criteria may not be isomorphic with that necessitated by syntax etc. Keeping these universally established guidelines, GEN can generate an innite set of candi-dates for each input. For instance, take the English word ‘grammar’. GEN can generate the following candidate representations and many more: [ǣф_̲_ф], [ǣŨф_̲_ф], [چǣф_څ̲_ф], [څǣф_چ̲_ф], [چǣŨф_څ̲_ф], [څǣŨф_چ̲_ф], [ǣŨچф_څ̲_ф]
and many, many more. These candidate representations generated accord-ing to some universal principles turn out to be non-optimal for English for well-founded reasons. But the point is that if the candidate set generated by GEN is large enough it will always include a unique, optimal candidate as well. And the idea of generating a large candidate set is to systematically rule out all logical possibilities other than the one actually chosen by the language.
When we turn to Carnatic music, we already saw that more than one interpretation is possible for an input and that more than one output may be optimal as far as the grammar is concerned. But due to paucity of research in music in general and Carnatic music in particular, we do not know the guiding principles of structure building in music. For instance, while some types of music may require structure similar to that of sentences in language partitioning the line into a subject/predicate type of bifurcation (Lerdahl and Jackendoff (1983), Gilbers and Schreuder (2002) for Western music), Car-natic music seems to require structuring only at the levels of tone, phrase and line, resembling the phonological hierarchy of segment, phonological word and intonational group (see chapter 4 and 5 for details). Further, we do not know the range of attested behaviour in the realms of pitch realisa-tions of notes, note boundaries, prominence in phrases, mapping rhythmic
Generator and the grammar of Carnatic music 59 prominence to phrasal prominence, etc. in music (see discussion at the end of chapters 5 and 7). Therefore, as we have no idea of the universal structural principles of music, GEN cannot generate an innite set of candidates for mu-sic and so we must be satised with a non-generative grammar of Carnatic music for the present.
But as we saw, even within the domain of Carnatic music we can, in fact, generate a fair number of candidate interpretations for every input musi-cal line. Of these plausible candidates, our grammar can even explain the ungrammaticality of a large majority of the candidates. But because of the nature of the Carnatic music lexicon, the explanation could be attributed to one of two types of FAITHLEX violations, namely a raagam-specic one or a
*AMBIGUITY type detected by the networked lexicon. As for the former, the constraint prohibiting pitch curve interpretations of two note sequences in the raagam Beegađaa is one such FAITHLEX constraint. Here is another ex-ample of a raagam-specic constraint pertaining to a prohibition on certain notes as boundary tones of musical phrases. The example is from the raagam Sahaanaa, the sample music demonstration in (12) in chapter 2 [2.24] given here as (24) below for ease of reference.
(24) Musical demonstration
Sahaanaa II [ 2.24]
Phrasing A: [ni sa] [ri gu ma pa][ di ni Sa]
Phrasing B: ni *[sa ri gu ma][ pa di ni Sa]
While phrasing A is grammatical, B is ungrammatical. Let us assume that the lexicon lists the information as a FAITHLEXSAHAANAA constraint given in (25).
(25) FAITHLEXSAHAANAA
FAITHLEXSAHAANAA[sa → *ma] ( informally, an ascending phrase initiated by the note sa cannot have ma as the terminal note) When an input string {ni sa ri gu ma...} is parsed as ...ni [sa ri gu ma]... then FAITHLEXSAHAANAA is violated and the output is deemed to be ungrammatical though other parses like [ni sa][ri gu ma], [ni sa ri] [gu ma] and [ni sa ri gu] ma and several others will not.
Turning to the second type of ungrammaticality, it is seen that it can be attributed to the constraint family *AMBIGUITY. Consider the instance of
un-60 Issues in modeling the grammar
grammaticality, once again from a musical line in the raagam Beegađaa. The music demonstration [6.59] reproduces a musical line the author had come across in a live concert of a highly acclaimed artist. ((11–13) of chapter 6 and music demonstrations [6.59]–[6.62] reproduced below as (26); see chapter 6 for a detailed discussion).
(26) Musical demonstration
Beegađaa [ 6.59]
Varͣam: Beegađaa: Aadi: Viiͣai Kuppayyer The text: // pa, di pa - ma; pa / ; ma pa - di,...
bha ga va: ri
*// pa di Sa, Ri Sa - ni di ni di pa, /; ...
bo: da na
Beegađaa [ 6.60]
Di pa Sa Sa Ni; di pa
Kaamboodi I [ 6.61]
Pa di Sa Sa Ni di pa
Varͣam II (contd.) [ 6.62]
(?)// pa di Sa, Ri Sa - ni; di / pa
The reason for the ungrammaticality is that the musical line in the raagam Beegađaa smudges the raagam boundary between the raagam Beegađaa and another raagam, Kaamboodi. I assume that a major function of the Carnatic music lexicon is to create a network of raagas so that raagam identities are kept clear at all points of the discourse (see chapter 8 for a detailed discuss-sion). In other words, creating an entire musical line ambiguous between two raagas is not just a matter of bad taste, it is a matter of ungrammaticality. I term the constraint *AMBIGUITY (*AMB for short) as this a general constraint type which signals the smudging of raagam boundaries dened explicitly in the networked lexicon of Carnatic music (see chapters 5, 6, 7, and 8). In the case under discussion, the opening phrase in the rendering [pa di Sa, RiSa]
will automatically signal *AMB for the raagam Kaamboodi and the situation gets worse with the next phrase with the short note ni in the descent which is not sanctioned for Beegađaa but only for Kaamboodi and the line earns a serious *AMB violation for the constraint *[Sa [*ni di] which is part of the FAITHLEX constraints specic to the raagam Beegađaa making the line ungrammatical.
EVAL and the grammar of Carnatic music 61 The family of *AMB constraints, in addition to assigning a violation mark for a specic constraint in the set, identify network connections which trace the cause of the violation in the FAITHLEX constraint set of one raagam to an-other raagam/an-other raagas. In this case, the phrase Sa [ni di...] triggers a vio-lation of the Beegađaa specic constraint *[Sa [*ni di... and the *AMB family of constraints trace the cause of the ungrammaticality to another raagam in the network. There is enough anecdotal evidence in Carnatic music circles of listeners actually identifying the raagas which have crept in inadvertently during the rendering of a raagam by less experienced performers. (See chap-ters 5, 6, 7, and 8 for discussions of *AMB constraints).
Getting back to our discussion of the role of GEN in the grammar of Car-natic music, though, in principle, every CarCar-natic musician of standing will be able to generate a fair number of plausible interpretations of a notated line of music constituting the input (if he/she can read music notation), we still do not have a clear idea of the limits of musical structure that should be consid-ered for generating serious contenders for evaluation. This is the reason why the model of grammar that I propose for Carnatic music is not a ‘generative’
one as it lacks the component GEN. Therefore the model that I propose scans a musical line, assigns it structure and with the help of relevant FAITHLEX constraints evaluates it for a) grammaticality (chapters 4 and 5) and b) musi-cal content (chapters 6 and 7).
However, the question of only one of the interpretations, i.e. a unique candidate representation being considered optimal, remains. As I said earlier, the ability to offer multiple interpretations for a line of music is an inherent characteristic of Carnatic music discourse (as the discourse is not meant to be a reproduction of a ‘xed’ text but a spontaneous creation of musical ideas (see the last section of chapter 9)), I do not take on myself the task of parallel evaluation of multiple candidate representations at this stage of musical research. The model of grammar for Carnatic music is thus not quite an OT model as it is non-generative and it lacks the power to select a unique candidate as the optimal one from among a set of candidates.