On Greek, Ancient and Modern (but Rather Ancient than Modern)

In this paper a few applications of the current implementation of R are given. These applications concern typesetting problems that cannot be solved by TEX (consequently, by no other typesetting system known to the authors). They cover a wide range, going from calligraphic script fonts (Adobe Poetica), to plain Dutch/Portuguese/Turkish typesetting, to vowelized Arabic, fully diacriticized scholarly Greek, or decently kerned Khmer. For every chosen example, the particular difficulties, either typographcal or TEXnical (or both), are explained, and a short glance to the methods used by ! 2 to solve the problem is given. A few problems Q cannot solve are mentioned, as challenges for future Cl versions. On Greek, ancient and modern (but rather ancient than modern) Diacritics against kerning. It is in general expected of educated men and women to know Greek letters. Already in college, having used 0 for angles, y for acceleration, and n to calculate the area of a round apple pie of given radius, we are all familiar with these letters, just as with the Latin alphabet. But the Greek language, in particular the ancient one, needs more than just letters to be written. Two lunds of diacritics are used, namely accents (acute, grave and circumflex), and breathings (smooth and raw) which are placed on vowels; breathings are also placed on the consonant rho. Every word has at most one accent1, and 99.9% of Greek words have exactly one accent. Every word starting with a vowel has exactly one breathing2. It follows that writing in Greek involves much more accentuation than any Latin-alphabet language, with the obvious exception of Vietnamese. How does TEX deal with Greek diacritics? If the traditional approach of the \accent primitive had been taken, then we would have practically no hySometimes an accent is transported from a word to the preceding one: &vOpwn6q T L ~ instead of &vOpwnoq, ziq, so that typographically a word has more than one accent. 'With one exception: the letters pp are often written $6, when inside a word: no$bG. phenation (wbch in turn would result in disastrous over/underfulls, since Greek can easily have long words like cjtop~volapuyyoloy~x~), no kerning, and a cumbersome input, involving one or two macros for every word. The first approach, originated by Silvio Levy (1988), was to use TEX'S ligatures ('dumb' ones first, 'smart' ones later on) to obtain accented letters out of combinations of codes representing breathings (> and <), accents ( ' , ' and or =) and the letters themselves. In this way, one writes > ' h to get 4. Thls approach solved the problem of hyphenation and of cumbersome input. Nevertheless, t h s approach fails to solve the kerning problem. Let's take the very common case of the article zb (letter tau followed by the letter omicron); in almost all fonts there is a kerning instruction between these letters, obviously because of invariant characteristics of their shapes. Suppose now that omicron is accented, and that one writes t ' o to get tau followed by omicron with grave accent. What TEX sees is a 't' followed by a grave accent. No kerning can be defined for these two characters, because we have no idea what may follow after the grave accent (it can very well be a iota, and usually there is no kerning between tau and iota). When the letter omicron arrives, it is too late; TEX has already forgotten that there was a tau before the grave accent. A solution to this problem would be to write diacritics after vowels ("post-positive notation"). But 344 TUGboat, Volume 15 (1994), No. 3 -Proceedings of the 1994 Annual Meeting First applications of Q: Greek, Arabic, Khmer, Poetica, IS0 1 0 6 4 6 / ~ ~ 1 c O ~ ~ , etc. t h s contradicts the visual characteristics of diacritics in the upper case, since these are placed on the left of uppercase letters: "Eorp could hardly be transliterated E>'ar. And, after all, TEX should be able to do proper Greek typesetting, however the letters and diacritics are input. R solves this problem by using an appropriate chain of R Translation Processes (QTPs), a notion explained in Plaice (1994): as example, consider the word Zap: 1. Suppose the user wishes to input hs/her text in 7-bit ASCII; he/she will type > ' ea r , and this is already ISO-646, so that no particular input translation is needed. Another choice would be to use some Greek input encoding, such as rso8859-7 or EAOT; then he/she may as well type > ' Eap or >Cap (the reason for the absurd complication of being forced to type either > ' E or >t to obtain Z, is that "modern Greek" encodings have taken the easy way out and feature only one accent, as if the Greek language was born in 1982, year of the hasty and politically motivated spelling reform). The first RTP will send these codes to the appropriate 16-bit codes in IS0 ~ O ~ ~ ~ / U N I C O D E : Oxlf14 for Z, 0 ~ 0 3 b l for a, and 0 x 0 3 ~ 1 for p. 2. Once R knows what characters it is dealing with (Qrs default internal encoding is precisely rso10646), it will hyphenate using 16-bit patterns. 3. Finally, an appropriate QTP wdl send Greek IS