Proceedings of First International Symposium on Sanskrit Computational Linguistics

The most authoritative description of the morphophonemic rules that apply at word boundaries (external sandhi) in Sanskrit is by the great grammarian Pān. ini (fl. 5th c. B. C. E.). These rules are stated formally in Pān. ini’s grammar, the As. t .ādhyāyı̄ ‘group of eight chapters’. The present paper summarizes Pān. ini’s handling of sandhi, his notational conventions, and formal properties of his theory. An XML vocabulary for expressing Pān. ini’s morphophonemic rules is then introduced, in which his rules for sandhi have been expressed. Although Pān. ini’s notation potentially exceeds a finite state grammar in power, individual rules do not rewrite their own output, and thus they may be automatically translated into a rule cascade from which a finite state transducer can be compiled. 1. SANDHI IN SANSKRIT Sanskrit possesses a set of morphophonemic rules (both obligatory and optional) that apply at morpheme and word boundaries (the latter are also termed pada boundaries). The former are called internal sandhi (< sam. dhi ‘putting together’); the latter, external sandhi. This paper only considers external sandhi. Sandhi rules involve processes such as assimilation and vowel coalescence. Some examples of external sandhi are: na asti > nāsti ‘is not’, tat ca > tac ca ‘and this’, etat hi > etad dhi ‘for this’, devas api > devo ’pi ‘also a god’. This work has been supported by NSF grant IIS-0535207. Any opinions, findings, and conclusions or recommendations expressed are those of the author and do not necessarily reflect the views of the National Science Foundation. The paper has benefited from comments by Peter M. Scharf and by four anonymous referees. The symbol 〈’〉 (avagraha) does not represent a phoneme but is an orthographic convention to indicate the prodelision of an initial a-. 2. SANDHI IN Pān. ini’S GRAMMAR Pān. ini’s As. t .ādhyāyı̄ is a complete grammar of Sanskrit, covering phonology, morphology, syntax, semantics, and even pragmatics. It contains about 4000 rules (termed sūtra, literally ‘thread’), divided between eight chapters (termed adhyāya). Conciseness (lāghava) is a fundamental principle in Pān. ini’s formulation of carefully interrelated rules (Smith, 1992). Rules are either operational (i. e. they specify a particular linguistic operation, or kārya) or interpretive (i. e. they define the scope of operational rules). Rules may be either obligatory or optional. A brief review of some well-known aspects of Pān. ini’s grammar is in order. The operational rules relevant to sandhi specify that a substituend (sthānin) is replaced by a substituens (ādeśa) in a given context (Cardona, 1965b, 308). Rules are written using metalinguistic case conventions, so that the substituend is marked as genitive, the substituens as nominative, the left context as ablative (tasmāt), and the right context as locative (tasmin). For instance: 8.4.62 jhayo ho ’nyatarasyām jhaY-ABL h-GEN optionally This rule specifies that (optionally) a homogenous sound replaces h when preceded by a sound termed jhaY — i. e. an oral stop (Sharma, 2003, 783–784). Pān. ini uses abbreviatory labels (termed pratyāhāra) to describe phonological classes. These labels are interpreted in the context of an ancillary text of the The traditional classification of rules is more fine-grained and comprises sam. jñā (technical terms), paribhās. ā (interpretive rules), vidhi (operational rules), niyama (restriction rules), pratis. edha (negation rules), atideśa (extension rules), vibhās. ā (optional rules), nipātana (ad hoc rules), adhikāra (heading rules) (Sharma, 1987, 89). Proc. of FISSCL, Paris, October 29-31, 2007 As. t .ādhyāyı̄, the Śivasūtras, which enumerate a catalog of sounds (varn. asamāmnāya) in fourteen classes (Cardona, 1969, 6):