Devanāgarī is not an alphabet. It is an abugida, where every consonant already carries the vowel a - क is not k but ka. Other vowels are written by attaching a mātrā to that consonant, and the inherent a is cancelled with a virāma (्) when a consonant has to stand bare. This is why counting letters gives a misleading picture of the work: there are only about forty-eight signs, but they combine.

Letters hang from the śirorekhā, the horizontal head line running along the top, which is what gives a line of Devanāgarī its distinctive banner. Vowels have two forms - an independent one for the start of a word (अ, इ, उ) and a mātrā for everywhere else (ा, ि, ु). The vowel sign ि is worth meeting early: it is written before the consonant it follows in speech, so the eye and the ear disagree until the pattern settles.

When consonants meet with no vowel between them they fuse into conjuncts - क् + ष becomes क्ष, and ज् + ञ becomes ज्ञ. A few hundred of these exist; perhaps thirty are common. They are the part that looks impossible at first and turns out to be the part that makes reading fast, because a conjunct is recognised as one shape rather than assembled letter by letter.

The script is shared, not exclusive. Hindi, Marathi and Nepali use it too, so the letter shapes transfer directly and material written for any of them is useful here. Sanskrit, meanwhile, has been written in many scripts across its history and is perfectly at home in roman transliteration - which is why most of this site uses IAST, the scholarly romanisation that writes ā ī ū ṛ ṅ ñ ṭ ḍ ṇ ś ṣ ṃ ḥ rather than the script itself.

IAST and ISO 15919

Two romanisations are in common use and they disagree in a handful of places. IAST - the International Alphabet of Sanskrit Transliteration, settled at the Orientalist Congress in Geneva in 1894 - was built for Sanskrit and nothing else, which is why it is compact and why a century of scholarship is written in it. ISO 15919, standardised in 2001, was built to romanise every Indic script - Devanāgarī, Bengali, Tamil, Telugu, Kannada - with one Latin form per character, so that text can be turned back into the original without guesswork.

They agree on most of it: ā ī ū for the long vowels, ṭ ḍ ṇ ṣ for the retroflexes, ś for the palatal sibilant, ṅ ñ for the velar and palatal nasals, ḥ for visarga. They part company over the anusvāra (ं), which IAST writes ṃ and ISO writes ṁ; over the vocalic r and l (ऋ ऌ), IAST's ṛ ḷ against ISO's r̥ l̥; and over ए and ओ, which ISO writes ē ō so it can keep e and o free for the short vowels that Dravidian languages distinguish and Sanskrit does not.

The disagreement over the anusvāra is not arbitrary. In the consonant series a dot below is not decoration - it marks retroflex articulation. ट ठ ड ढ ण ष become ṭ ṭh ḍ ḍh ṇ ṣ, every one of them made with the tongue curled back against the hard palate, as against the dentals त थ द ध न written t th d dh n. A dot below is a claim about where the tongue goes.

Which is the case against ṃ. The anusvāra is not retroflex and has nothing to do with curling the tongue back: it is nasalisation of the vowel before it, or a nasal that takes its place of articulation from whatever follows. Writing it with the mark that everywhere else means retroflex files it in a series it does not belong to. Moving the dot above - oṁ, saṁskāra, evaṁ - leaves the dot below with exactly one job, which is why a good many people who otherwise write IAST make this one substitution. The visarga keeps its dot below in both systems, a rough edge neither standard chose to sand down.

The same reasoning explains ISO's r̥. IAST's ṛ borrows the retroflex dot for a sound that is not a retroflex consonant at all but a vowel - r doing a vowel's work in the middle of a syllable, as in kṛṣṇa. A ring below is the standard mark for a syllabic consonant, so r̥ says this r is behaving as a vowel rather than this r is retroflex. The ambiguity is not hypothetical: in romanised Hindi ṛ is usually ड़, the retroflex flap, a different sound entirely.

This site writes IAST, with ṃ for the anusvāra. Expect to meet both: the same word can appear as saṃskāra or saṁskāra, and as kṛṣṇa or kr̥ṣṇa, without either being wrong.