English to Sanskrit

Sanskrit is an ancient language of the Indian subcontinent, estimated to date back to the second millennium BCE or earlier. It is the language of the religious texts, liturgy, and the great epics of several religious orders in South Asia. The rules of the Sanskrit language were codified by Panini in his monumental work, the 'Ashtadhyayi', estimated by historians to have been written between the fifth and the third century BCE. While the language is no longer spoken by the masses of people, Sanskrit grammar has enjoyed a central place in the traditional Indian schools for novitiates for several thousand years. The Ashtadhyayi has been taught to novitiates for well over 2000 years ! It describes both the language used in the early religious texts and liturgy (Vedic Sanskrit) as well as that used in the great epics and literature (Classical Sanskrit).

In this first release of our English to Sanskrit Machine Translation system, we have excluded certain elements of Classical Sanskrit that were not considered essential in a translation from English to Sanskrit. One such case is the Aorist past tense in Sanskrit. The Past tense in Sanskrit has three forms - the imperfect past, the Aorist past, and the perfect past. However, experts have noted that they have been used interchangeably in much of historical literature, the distinctions between these three forms having become fuzzy over time. On a more practical note, it is not possible for the machine to identify them accurately in the absence of clear rules and 'knowledge of the world'. Hence we have excluded the Aorist past tense in its entirety. Certain other structures were also excluded as they are not essential for translation from English to Sanskrit (and some are relatively rare forms); they can be expressed in alternate, possibly simpler structures (but perhaps with some loss of brevity). These include certain kinds of participial forms, periphrastic future, some verb derivative forms (causative, benedictive, desiderative, and frequentative), compound terms, and adjectival phrases formed from subordinate clauses (for e.g., 'kind-hearted man' instead of 'the man whose heart is kind'). [1][2][3][4][5][6][7][8]

A sample translation is shown on the next page for a sentence that contains a few interesting syntactic structures (relative clause, subordinate infinitive clause, negation, adjunct preposition phrase).

The first release of English to Sanskrit MT includes the following features:

Infinitives have largely been implemented in the narrower sense assumed in Sanskrit (but a future revision may treat all infinitives in a uniform manner, in order to avoid the use of rare forms found only in Vedic Sanskrit). On the other hand, the 'past perfect' has been implemented in the sense assumed in most other languages including English (e.g. 'he had finished the job'), rather than the sense in which the term is defined in Sanskrit (i.e. past events unwitnessed by the speaker). However, the 'present perfect' (e.g. 'he has finished the job') is treated in the same way as the 'past imperfect', and is thus differentiated from the 'past perfect'. In Classical Sanskrit, the 'present perfect' is probably best treated in the Aorist form (but the Aorist has not been implemented in this version of the software). This treatment of the 'perfect' may not be problematic, given the interchangeability of past tense forms discussed previously.

Sanskrit is much more complex (in a computational sense) than most of the Indian languages derived from it for a number of reasons. These include the additional Gender (Neuter), the additional Number (Dual), the additional 'first-class' Cases (3 to 8) attached to Nouns and Adjectives, the complexity associated with postpositions (and specific Case assignment and agreement), the complex verb terminations for different tenses/ aspects/ moods (with the added complexity from Strong/ Weak and Class/Voice distinctions), and the exceptions to various rules. One measure of its comparative complexity is the number of 'descriptive rules' in the target language generation component of an MT system. Ideally, the language generation code should be very compact, and should be driven by the 'descriptive rules' for each language, that describe elements such as verb conjugation, declension, and 'fusion' of morphemes. At this point of time, Classical Sanskrit has 4x the number of 'descriptive rules' compared to Hindi in this MT system. And these rules do not yet include the complexity of Vedic Sanskrit !

Machine Translation to Sanskrit has been particularly challenging to implement and has taken considerable time and effort. However this effort has also been extremely rewarding.

References