Relative efficiencies of devanagari and tamil scripts for reading

Abstract

This thesis is devoted primarily to some of the problems involved in reading with particular emphasis on Devanagari and Tamil scripts. Scripts are considered as codes and their performance examined on the basis of information they can convey in a given time. Specifically, when different linguistic expressions are treated as alternative codes to express the same ideas, it is required to estimate the representation in which more information can be conveyed in a given time or in a given number of symbols. Thus assuming that the same idea is encoded and represented in the three scripts, namely, English, Tamil and Devanagari, it is required to find the representation in which information can be taken in faster in reading. The work reported in this thesis was inspired by and is based on that of Miller who had established that the number of letters perceived during brief exposures is essentially limited by the memory span. He found that when a string of English characters is exposed to a subject, he can perceive and retain a certain number of letters, which he can report immediately by writing. In this study, this invariance in the memory span is examined independently for Tamil and Hindi and it is shown that the number of bits or graphemes perceived is constant in both character representation and in syllable representation. The invariance in memory span is extended to different languages on the basis of number of graphemes perceived in one fixation. The average entropy per syllable is estimated in the two languages to compare the information intake with syllable characters in any one given language. To find the number of graphemes perceived, the tabulation of the entropies for the various orders of approximation is required and so they were determined using Shannon’s Predictive Entropy. To estimate the relative efficiency of different scripts in reading a normal text in different languages, the predictive entropy of the hundredth grapheme is used. This entropy can be used to estimate the actual number of graphemes perceived in reading long sequences of messages. A comparison based on the total number of bits perceived in different languages having the same semantic content shows that Indian languages are more efficient. The scheme of writing in which each speech sound is represented by its basic character is shown to be less efficient than the conventional syllabic representation. The problem of reading Indian languages written in English alphabet, using a certain transliteration scheme, is also investigated. The performance of transliteration is shown to be poorer than that of the conventional syllable representation.