Encoding & Transliteration

CDSL stores Sanskrit text internally in SLP1, and lets you input and display words in several transliteration schemes. Knowing which scheme you are typing in is the single most common source of "why didn't my word match" confusion.

Schemes the displays offer

SLP1 is the website's lingua franca (storage encoding). The display input/output menus let you read or type in these schemes (the transcoder names are the values the API uses):

Scheme (transcoder name)	What it is	rāma	śiva
slp1	Sanskrit Library Phonetic Basic — the internal encoding; one ASCII char per phoneme	rAma	Siva
deva	Devanāgarī (Unicode)	राम	शिव
roman	Roman with diacritics (IAST-style)	rāma	śiva
hk	Harvard-Kyoto — ASCII, case-sensitive	rAma	ziva
itrans	ITRANS — ASCII scheme	rAma	shiva

Why two example words

rāma looks identical in SLP1, Harvard-Kyoto, and ITRANS (all rAma) — none of its sounds are written differently across those ASCII schemes, so for this word the input scheme barely matters. śiva is the opposite: the ś sound is S in SLP1, z in Harvard-Kyoto, and sh in ITRANS — three different ASCII spellings of the same letter. The sounds where the schemes actually diverge are the sibilants (ś, ṣ), vocalic ṛ/ḷ, and the retroflex consonants. When your word contains one of those, choosing the right input scheme is exactly what makes the lookup match.

Two more transcoders exist but are not in the display input/output menus: wx (used at Hyderabad University) and as (Anglicized Sanskrit, Thomas Malten's letter-number scheme). slp1 and deva are the values seen in the API input/output parameters.

Why SLP1 internally

SLP1 is lossless and unambiguous: every Sanskrit phoneme maps to exactly one ASCII character, so sorting, searching, and round-tripping are deterministic. The downloadable XML is in SLP1 for this reason (see Downloads & Data).

Vedic accents

Several dictionaries record Vedic pitch accents, and the data carries them — for Monier-Williams alone the accent mark sits on tens of thousands of headwords.

Which dictionaries mark accents:

Dictionary	Accents?
Monier-Williams (MW)	Yes
Böhtlingk-Roth (PWG) and shorter (PW)	Yes
Grassmann (GRA)	Yes — central; it is a Ṛg-Veda dictionary
Apte (AP90)	No

How they are encoded. In the SLP1 source the udātta (high pitch) is written / attached to its vowel — e.g. MW keys agni as agni/ (accent on the final -i, agní). The other Vedic accents (anudātta, svarita) appear as \ and ^ in the entry body of the German and Grassmann sources. When rendered, accents show in Roman as a combining acute (á, ā́) and in Devanāgarī as the usual udātta/anudātta strokes. The precise marking scheme is described per dictionary on its csl-doc page.

You never have to type an accent

Each entry stores a plain search key and a separate accented display key (<k1>agni vs <k2>agni/). Lookup uses the plain key, so typing agni finds the word whether or not you know its accent. The accent is for display, not for matching.

Showing or hiding accents. The displays offer an accent toggle, and the API exposes it as accent=yes|no (it governs whether accents appear in Devanāgarī output). Turn it off for a cleaner reading text; on to study the pitch.

Typing a headword

• If you have a diacritic Roman word (rāma), choose the roman input mode.
• If you only have an ASCII keyboard, Harvard-Kyoto (hk), ITRANS (itrans), or SLP1 (slp1) avoid diacritics entirely.
• If you have Devanāgarī, paste it directly and select the deva input mode.

If a lookup fails, the most likely cause is an input-scheme mismatch (e.g. typing roman while the box expects hk). Re-check the selected scheme first.

Transcoding tooling

Scheme conversion uses the project's transcoder, developed in Java by Ralph Bunker and ported to PHP (transcoder.php) and Python (funderburkjim/sanskrit-transcoding). It works from per-pair XML tables named {X}_{Y}.xml (e.g. slp1_deva.xml), compiled into a finite-state machine at runtime. For example transcoder_processString('rAma', 'slp1', 'deva') returns राम. See also Data Formats.