Note: the following link gives online access to the great Monier-Williams Sanskrit-English Dictionary: http://webapps.uni-koeln.de/tamil/
For the purposes of this discussion, the following system of transliteration is used to represent the Sanskrit alphabet:
- a A i I u U R RR L LL
- e ai o au
- M [anusvAra]
- : [visarga]
- k kh g gh n;
- ch chh j jh n'
- T Th D Dh N
- t th d dh n
- p ph b bh m
- y r l v
- sh Sh s
[Note that proper nouns are not capitalized in this system. The accent in words like jn'Ana can be readily omitted [= jnAna], as likewise with an;ga [=anga]. The ungainly chchh can conveniently be written cch.]
Sanskrit spelling is perfectly phonetic. At the same time, the pronunciation of many letters is affected when they come in contact with particular letters. For this reason, the perfectly phonetic spelling creates inconsistencies. 'Undoing' these inconsistencies is the first task of a translator.
The Sanskrit alphabet contains 42 letters [varNas]: 9 vowels [sveres] and 33 consonants [vyanjanas].
There are 5 simple svaras [a i u R L], and four diphthongs [e ai o au].
A vowel may be short [hrasva], long [dIrgha], or protracted [pluta], according to its length. The lengths are respectively 1, 2, and 3 beats [mAtras]. Thus:
- a i u R L
- A I U R e ai o au [Note that the letter LL is merely hypothetical].[A short vowel is considered to be long in poetry when followed by more than one consonant, or by a final consonant.]
- a3 A3 i3 I3 u3 U3 R3 RR3 L3 e3 ai3 o3 au3
[The plutas are encountered only in chant, in a voice calling from a distance, or in echo].
Vowels may be sounded with or without nasalization [anunAsika or ananunAsika], as in French. Nasalization is indicated by the letter M, a form called anusvAra.
Vedic Sanskrit also distinguishes tones of vowel, but they are disregarded in Classical Sanskrit. The tones or accents are acute [udAtta], grave [anudAtta], and circumflex [svarita].
The 33 consonants are grouped, according to their articulation, as mutes [sparshas], intermediate semivowels [antasthas], and sibilants [Ushmans]. The sparshas are further grouped into five vargas, according to the place of articulation. These are the vyan'janas:
- [a] Mutes:  k, kh, g, gh, n; [guttural];  ch, chh, j, jh, n' [palatal];  T Th, D, Dh, N [retroflex lingual or 'cerebral'];  t, th, d, dh, n [dental];  p, ph, b, bh, m [labials]
- [b] Semivowels: y, r, l, v
- [c] Sibilants: sh, Sh, s
- [d] Sonant aspirate: h.
The first two letters in each of the 5 vargas [k, kh, ch, chh, etc.] are called surds or harrd consonants. The other three in each varga are sonants or soft consonants.
Visarga is a hard breathing that is badly pronounced by modern speakers. It is, in fact, a vestigial sound, like the letter 's' or 'r' when the tongue does not quite contact the roof of the mouth a whispered shh. It never has the sound of hah, a pronunciation that violates the meter of the verse in which it is read.
Consonants are with or without breath [mahAprANa or alpaprANa]. The 1st and 3d letters of each varga, together with the 5th [nasals], and the semivowels, are alpaprANa, the rest are mahAprANa.
The vargas are based on the locus of articulation. The gutturals [kaNTha] are formed in the throat; the palatals [tAlu] on the palate; the linguals with the tongue on the roof of the mouth [mUrdhan]; the dentals on the teeth [danta]; and the labials on the lips [oShThau].
The avagraha sign is used to mark the elision of a and is here indicated by a single apostrophe. The initial a- is elided after final -e or -o, thus: te + api = te 'pi, them too; kAla: + asti = kAlo + asti = kAlo 'sti, it's time. When A follows A it is elided, and such elision may optionally be shown by the double avagraha, thus: tathA + Aste = tathA "ste, or tathAste, so be it.
Vowels have three degrees of weight: Simple, guNa, and vRddhi. For each of the vowels, these degrees are shown below:
- Simple: a / i, I / u, U / R, RR / L
- guNa: a / e / o / ar / al
- vRddhi: A / e / au / Ar / Al
Having these definitions in mind, we can proceed to the rules of sandhi.
[The following summary of the rules of sandhi is based on M.R. Kale’s grammar of 1894.] The term sandhi derives from sam ‘together’ and dhA ‘to join’. It indicates the coalescence of two letters on account of phonetic principles. == sandhi of vowels [svara-sandhi].
When a simple vowel, whether short or long, is followed by a similar simple vowel, short or long, the two vowels are replaced by the corresponding long vowel: for example:
- a + a = A, daitya + ari: = daityAri:, for of the children of Diti
- a + A = A, atra + AsIt = atrAsIt, here was
- A + a = A, yadA + abhavat = yadAbhavat, when he was
- A + A = A, vidyA + Atura:, eager for wisdom
- i + i = I, iti + iva = itIva, just so
- i + I = I, api + IkShate =apIkShate, indeed he wishes
- I + I = I, shrI + Isha: = shrIsha:, Shri's lord [Vishnu]
- u + u = U, bhAnu + udaya: = bhAnUdaya:, sunrise
- u + U = U, sAdhu + Uchu: = sAdhUchu:, they said well
- U + U = U, chamU + Urja: = chamUrja:, the strength of an army
- R + R = RR, kartR + Rju: = kartRRju:, doer of right
- R + RR = RR, kR + RRkAra = kRRkAra:, the letter RR with the root kR
- R + L = RR, hotR + LkAra: = hotRRkAra:, the letter L as pronounced by the priest
When different simple vowels are combined, the corresponding guNa letter replaces them, e.g.:
- a + i = e, upa + indra: = upendra:, Indra's younger brother [Vishnu]
- a + I = e, parama + Ishvara: = parameshvara:, the supreme Lord
- A + i = e, ramA + ichchhA = ramechchhA, the wish of the lady Rama
- A + I = e, yathA + Ipsitam = yathepsitam, as desired
- a + u = o, hita + upadesha: = hitopadesha:, friendly instruction
- a + U = o, kRShNa + Uru: = kRShNoru:, Krshna's thigh
- A + u = o, gan;gA + udakam = gan;godakam, Ganges water
- A + U = o, mahA + Uru: = mahoru:, great thigh [?]
- a + R = ar, kRShNa + Rddhi: = kRShNarddhi:, the prosperity of Krshna
- A + R = ar, mahA + RShi: = maharShi:, great sage
- a + L = al, tava + LkAra: = tavalkAra:, the 'L' you say
Sometimes when similar consonants meet, one is optionally dropped in the process of sandhi, e.g.: kRShNa + Rddhi: = kRShNarddhi:, or kRShNar + d + dh + i: = kRShNardhi:, where the similar d is dropped. There is also a third form kRShnarddhi: [see below].
When any consonant of the five vargas [except the palatal nasal n'] follows a semivowel [y r l v], it is optionally doubled; so the above example tavalkAra: may also be tavalkkAra:, tavallkAra:, or even tavallkkAra:
- When a word ending in -a is followed by Uha; or when pra is followed by Uha, UDha, or UDhi, e.g.: praShTha + Uha: = praShThauha:, chief guess; pra + Uha: = prauha:, principal reasoning.
- akSha + UhinI = akShauhiNi [for the change from n to N see below].
- When sva is followed by Ir or Irin [from the root >Ir to go], e.g.: sva + Ira: = svaira:, a wilful person.
- When -a is followed by Rta with an Instrumental sense, e.g.: sukha + Rta: = sukhArta:, blessed with happiness.
- Some compounds with -RNa, e.g.: pra + RNam = prArNam, principal debt; RNa + RNam = RNArNam, a second mortgage.
- When a preposition ending in -a precedes a > root beginning with R-, as upa- + Rchchhati = upArchchhati.
When a or A is followed by e or ai and o or au, then ai and au are the substitutes, e.g.: kRShNa + ekatvam = kRShNaikatvam, oneness with Krshna; deva + aishvaryam = devaishvaryam, divine majesty; sA + eva = saiva, she alone; bhava + auShadham, medicine against reincarnation; vidyA + autsukyam = vidyautsukyam, longing for wisdom.
- When a preposition ending in -a precedes a >root beginning either e- or o-, then e or o is the substitute, e.g.: pra + ejate = prejate, quakes severely.