Nepali typing reference

SahajLipi currently converts Roman Nepali, 51 listed English-spelling loanword stems with finite attached suffix forms and 12 full English month names to Unicode Devanagari. The core returns a preferred reading and, where listed, alternatives. The optional browser adapters render that preferred reading in opted-in text fields as you type and report alternatives to the host interface. This reference describes the published 0.1.0-alpha.4 package, including the earlier 13-key spelling slice and 34 native forms released on 2026-10-09; the keys are project conventions rather than a standardized Romanization scheme. The earlier 0.1.0-alpha.1 keeps its half-consonant default; see the migration recipe.

The starter lexicon takes priority over the phonetic fallback. The listed word cha defaults to च and offers छ as an alternative; chha returns only छ. The fallback tokens remain ch → च and chh → छ. Custom entries can replace a built-in entry in one engine instance. See the engine source for the lookup order.

Vowels

A bare or final consonant is full by default: both k and ka → क. The vowel a supplies no visible vowel sign and separates consonants where needed: kar → कर, while kr → क्र. Longer vowels are typed explicitly, so pani → पनि and paani → पानी.

The table shows independent vowels and examples after k. The R row is case-sensitive.

Roman keys	Independent vowel	After k
a	अ	ka → क
aa	आ	kaa → का
i	इ	ki → कि
ii or ee	ई	kii or kee → की
u	उ	ku → कु
uu or oo	ऊ	kuu or koo → कू
e	ए	ke → के
ai	ऐ	kai → कै
o	ओ	ko → को
au	औ	kau → कौ
R	ऋ	kR → कृ

Lowercase ri → रि and kri → क्रि. Use capital R when you mean ऋ or the ृ vowel sign.

Consonants

These are fallback tokens, shown as their base letters. A bare or final token is full by default: k → क and kh → ख; ka and kha give the same letters. Adjacent consonants form internal conjuncts automatically. Longer tokens are matched before shorter ones. A listed word can override the fallback result.

Roman token	Base letter	Roman token	Base letter
k	क	kh	ख
g	ग	gh	घ
ng	ङ	c or ch	च
chh	छ	j	ज
jh	झ	ny	ञ
T	ट	Th	ठ
D	ड	Dh	ढ
nn	ण	t	त
th	थ	d	द
dh	ध	n	न
p	प	ph or f	फ
b	ब	bh	भ
m	म	y	य
r	र	l	ल
v or w	व	s	स
sh	श	S or Sh	ष
h	ह	ksh	क्ष
gy	ज्ञ

Shift changes a few sounds

Lowercase t and d produce dental sounds; capital T and D produce retroflex sounds. Keep h lowercase in the aspirated pairs.

Dental	Retroflex
ta → त	Ta → ट
tha → थ	Tha → ठ
da → द	Da → ड
dha → ध	Dha → ढ

Capital S or Sh selects ष (Sa or Sha → ष), while lowercase sh selects श (sha → श). Capital H after a vowel adds visarga ः: kaH → कः and duHkha → दुःख. Lowercase h remains ह (ha → ह), and : remains a colon. Other capitals become lowercase before lookup. The reserved sound keys T, D, S, R, and vowel-following H bypass lowercase word lookup; an exact cased entry can override them. For example, Saryo → षर्यो, gaRyo → गऋयो, and baHini → बःइनि follow explicit Shift input. The names Ram → राम and Sita → सीता and the month names September → सेप्टेम्बर and December → डिसेम्बर are explicit built-in aliases. Replacing their lowercase entries also updates those aliases unless you supply an exact cased override.

Half consonants and conjuncts

The default fallback displays full bare and final consonants immediately, while adjacent consonants form an internal cluster. Pressing Space or typing punctuation commits the displayed reading. Use backtick or / when the final consonant must remain half.

Type	Output	Why
k or ka	क	A bare final consonant is full
kr or kra	क्र	Internal क् joins full final र
kri	क्रि	i adds the ि sign
kar	कर	a separates क and र
shakti	शक्ति	Adjacent consonants form the cluster

Type backtick (`) or / to request a virama explicitly: k` or k/ → क्, and ka` or ka/ → क्. The explicit half form persists when you finish the word with a space or punctuation: k` followed by Space keeps क्. Each marker remains literal where no consonant can take a virama, such as a` → अ` or a/ → अ/, and the slash in 3/4 stays literal.

A vowel after either explicit half marker is independent: k`i and k/i → क्इ. Use ki → कि for the attached vowel sign.

Type backtick followed by = (`=) or /= after a consonant when you need a zero width joiner after its virama: par`=yo and par/=yo → पर्‍यो. The inserted sequence is virama U+094D followed by zero width joiner U+200D. The exact visible half form or conjunct depends on the browser's text shaping and font. See the Unicode Indic FAQ on half forms and joiners.

Since alpha.2, developers can select createEngine({ consonantMode: 'half' }) to retain alpha.1 phonetic fallback endings: k → क्, kr → क्र्, while ka → क and kra → क्र. The default is consonantMode: 'full'. This option changes fallback endings only; exact built-in and custom readings keep their priority and supplied Unicode. See the API option, browser migration recipe, and dated software-contract record.

र्य and र्‍य

The ordinary ry cluster uses र्य: र U+0930, virama U+094D, then य U+092F. The spelling र्‍य also includes zero width joiner U+200D between the virama and य to request the eyelash form of र. Unicode describes this mechanism in chapter 12, rule R5a. It does not specify which Nepali words should use it.

The starter lexicon supplies र्‍य for these exact listed words:

Roman word	Preferred output
paryo	पर्‍यो
garyo	गर्‍यो
maryo	मर्‍यो
maaryo	मार्‍यो
bharyo	भर्‍यो
taryo	तर्‍यो
saryo	सर्‍यो
puryaunu or puryaaunu	पुर्‍याउनु
bharyang or bharyaanga	भर्‍याङ

Title case follows the same entries, for example Garyo → गर्‍यो. The reserved Shift keys still select their explicit sounds; Saryo is therefore षर्यो, while lowercase saryo is सर्‍यो. Short and long vowels remain distinct: maryo → मर्‍यो and maaryo → मार्‍यो, just as pani → पनि and paani → पानी.

Two informal spellings have additional whole-word exceptions:

• puryaunu supplies आ followed by independent उ in पुर्‍याउनु. Its au does not use the ordinary औ token. puryaaunu spells the long vowel explicitly. The fallback still converts au to औ, as in kau → कौ.
• bharyang supplies the long आ in भर्‍याङ. bharyaanga explicitly includes aa and final a. Bare final consonants in the default fallback are full.

These entries match whole words; they do not infer stems, suffixes, or compound spellings. Other words keep the ordinary cluster, including kaarya → कार्य, suurya → सूर्य, saundarya → सौन्दर्य, and aachaarya → आचार्य. For an unlisted form that needs र्‍य, type /= explicitly and supply its vowels: gar/=yo → गर्‍यो, pur/=yaaunu → पुर्‍याउनु, and bhar/=yaanga → भर्‍याङ.

Online usage supports the listed Nepali spellings: नेपाल कानून पत्रिका uses गर्‍यो and मर्‍यो and मार्‍यो; Nagarik uses भर्‍यो, Nepal magazine uses तर्‍यो, and Gorkhapatra uses सर्‍यो. The District Administration Office, Parsa uses पुर्‍याउनु in a notice heading, and नेपाल कानून पत्रिका uses भर्‍याङ. These are assisted checks of published spellings, not independent human verification of the Roman mappings; that review remains pending. The dated review and benchmark report records these decisions, checks, remaining ambiguities, and limitations.

Reviewed native-word spellings

Four earlier reviewed starter keys remain included in alpha.4:

Roman key	Preferred output
halyo	हाल्यो
nabhani	नभनी
gaunle	गाउँले
dindaina	दिँदैन

Each has one reading. The 2026-10-01 review records their sources and dated comparison.

Alpha.3 spelling update — 2026-10-09: twelve single-reading preferences and a second reading for bhagna bring this reviewed native-key group to 17 keys. The published sahajlipi@0.1.0-alpha.3 package includes this slice; the release record identifies the verified artifact. The 34 later native forms are shipped in alpha.4 and excluded from historical alpha.3.

Roman key	Alpha.3 output	Alternative
imandar	इमान्दार
sarasar	सरासर
sakos	सकोस्
kathanak	कथानक
arambha	आरम्भ
ekadhik	एकाधिक
jaghanya	जघन्य
pukar	पुकार
niskanda	निस्कँदा
bora	बोरा
utthan	उत्थान
samanjasya	सामञ्जस्य
bhagna	भग्न	भाग्न

bhagna keeps its existing displayed भग्न reading and offers the distinct भाग्न reading for selection. Its order preserves the earlier default; the frozen references specify no uniquely preferred meaning. The 2026-10-09 review records source access, decisions and unchanged-reference measurements.

Alpha.4 native-form batch — 2026-10-09: the published package adds another 34 complete keys. For example, futera → फुटेर, astitwalai → अस्तित्वलाई, karyaharuma → कार्यहरूमा, pairaheka → पाइरहेका and samhalnubhaeko → सम्हाल्नुभएको. The complete native-form guide lists every mapping, its prior output and source access. These additions are included in published alpha.4 and outside the historical alpha.3 artifact. See the release section and exact-version installation instructions.

Roman key	Displayed reading	Alternative
mahila	महिला	माहिला
shanta	शान्त	शान्ता
angrejharuko	अङ्ग्रेजहरूको	अंग्रेजहरूको

The maintainer chose mahila and shanta ordering; the explicit velar nasal comes first for angrejharuko to preserve the earlier fallback style. These are software preferences, not measured frequency rankings. Choose an alternative in the host interface or use Alt+2 while the word is active in an attached field. The exact forms apply in either consonant mode. This batch does not infer other native endings or add loanword stems; its 12 deferrals remain in evaluation with unchanged references.

These completed keys supply reviewed whole-word spellings. They preserve the vowel distinctions in the fallback: pani and paani remain distinct, and ki and kii retain their vowel lengths. The default fallback makes bare and final consonants full; explicit backtick or / keeps a consonant half. Explicit ^ / ~ still choose bindu / chandrabindu; Shift keys keep their explicit sounds.

Only the complete normalized key matches. Incidental title case such as Halyo or Imandar works; reserved sound capitals keep their phonetic meaning unless an exact custom entry exists. For example, Sarasar still starts with ष, and niSkanda retains its explicit ष sound. Attached native forms and misspellings are not inferred from these aliases. Developers can replace a complete key’s readings with createEngine({ entries }).

English-spelling loanwords

These 51 listed stems automatically use the Nepali loanword forms below. The browser adapters apply them in Nepali mode, and the core conversion functions use them directly. school keeps स्कुल first and adds स्कूल as an alternative; other built-in loanword stems currently have one reading. The suffix review records the new media key, school alternatives and finite attached forms. These are source-assisted project preferences, with independent human linguistic review pending. The original pilot and earlier expansion retain their dated counts and evidence.

English key	Preferred output	English key	Preferred output
camera	क्यामेरा	computer	कम्प्युटर
mobile	मोबाइल	phone	फोन
charger	चार्जर	printer	प्रिन्टर
mouse	माउस	internet	इन्टरनेट
email	इमेल	software	सफ्टवेयर
scanner	स्क्यानर	video	भिडियो
taxi	ट्याक्सी	bank	बैंक
cheque	चेक	file	फाइल
school	स्कुल; alternative स्कूल	college	कलेज
doctor	डाक्टर	nurse	नर्स

The original pilot above has 20 keys. The table below lists the 30 expansion entries plus the latest media stem:

English key	Preferred output
ambulance	एम्बुलेन्स
battery	ब्याट्री
bus	बस
car	कार
carpet	कार्पेट
company	कम्पनी
connector	कनेक्टर
courier	कुरियर
cricket	क्रिकेट
digital	डिजिटल
drone	ड्रोन
football	फुटबल
furniture	फर्निचर
hotel	होटल
inverter	इन्भर्टर
keyboard	किबोर्ड
laptop	ल्यापटप
microphone	माइक्रोफोन
motorcycle	मोटरसाइकल
office	अफिस
password	पासवर्ड
radio	रेडियो
restaurant	रेस्टुरेन्ट
router	राउटर
sofa	सोफा
telephone	टेलिफोन
ticket	टिकट
van	भ्यान
website	वेबसाइट
wifi	वाइफाइ
media	मिडिया

The media row is the latest addition after the 20-key pilot and 30-key expansion.

This spells a borrowed word rather than translating its meaning: school produces स्कुल rather than विद्यालय. mouse uses the computer-device reading. A developer can replace a key's full candidate list with the existing createEngine({ entries }) API.

Camera, Computer and Company match their lowercase entries. Reserved Shift keys still apply, so Doctor, School, CAMERA and COMPUTER are not blanket-lowercased aliases. The finite attached forms below inherit the stem reading; unsupported spellings such as cameramaa keep the fallback. Hyphenated e-mail is split by text conversion and is not the email alias.

Keep an entire English field literal by leaving it unattached, marking it data-sahajlipi-ignore, or disabling conversion. Text conversion and attached fields preserve recognized links, ASCII domain-shaped hosts, and ordinary ASCII email addresses by default: camera becomes क्यामेरा, while camera.com stays literal. Ordinary English phrases, code, and filenames still need explicit literal handling; camera_file and camera123 can still be converted. See mixed text.

Attached loanword forms

Append one of these lowercase keys directly to a listed English loanword stem:

Suffix key	Appended Nepali form	Example
ma	मा	schoolma → स्कुलमा / स्कूलमा
ko	को	camerako → क्यामेराको
ka	का	companyka → कम्पनीका
ki	की	companyki → कम्पनीकी
le	ले	companyle → कम्पनीले
lai	लाई	companylai → कम्पनीलाई
bata	बाट	companybata → कम्पनीबाट
sanga	सँग	mediasanga → मिडियासँग
mathi	माथि	companymathi → कम्पनीमाथि
haru	हरू	companyharu → कम्पनीहरू

haru may be followed by one of ma, ko, ka, ki, le, lai, bata, sanga or mathi: companyharumathi → कम्पनीहरूमाथि and schoolharuma → स्कुलहरूमा / स्कूलहरूमा. That makes 19 recognized suffix keys, not arbitrary repeated suffixes. Candidate order follows the stem; choose the school alternative through the existing dropdown or candidate API.

Only the 51 listed loanword stems participate. Months, native Nepali entries and arbitrary custom stem names do not gain suffix handling. Exact whole-word custom entries win; customizing a recognized stem changes its derived readings. Unsupported suffixes, cameramaa and unrecognized stems use ordinary conversion. Capitals reserved for Shift sounds and explicit marks keep their existing rules. The rule joins spelling components; it does not decide whether the form is grammatical in a sentence.

English month names

These 12 full English month names automatically use the listed Nepali spellings. Each has one candidate and works in core conversion and in attached fields while Nepali mode is on. Normal title case also works, including the exact aliases September and December. The month reference records the spelling evidence and scope; these are authorized source-assisted project preferences, with independent human linguistic review pending.

English key	Preferred output	English key	Preferred output
january	जनवरी	february	फेब्रुअरी
march	मार्च	april	अप्रिल
may	मे	june	जुन
july	जुलाई	august	अगस्ट
september	सेप्टेम्बर	october	अक्टोबर
november	नोभेम्बर	december	डिसेम्बर

Try January February March → जनवरी फेब्रुअरी मार्च and September December → सेप्टेम्बर डिसेम्बर. Other uses of reserved Shift keys still select sounds; there is no general folding of all-capital or mixed-case English. For example, SEPTEMBER and DECEMBER are not month aliases.

May and may both select the month spelling मे. The engine cannot distinguish the English modal verb “may” from the month name. Native Roman Nepali such as maya keeps its existing behavior. Unlisted abbreviations such as jan, feb and sep and attached forms such as januaryma do not inherit month entries. This feature spells Gregorian month names; it does not translate an English sentence or convert dates to Bikram Sambat. Recognized links and email addresses stay literal under the shared text policy; ordinary English phrases still require explicit literal handling.

Digits

In Nepali mode, ASCII number keys produce Devanagari digits by default:

Type	Output
0123456789	०१२३४५६७८९
3.14	३.१४
September 27, 2026	सेप्टेम्बर २७, २०२६
name123@example.com	name123@example.com
https://example.com:8080/a2?q=2026	https://example.com:8080/a2?q=2026

The decimal point, commas, signs, slashes, and spaces keep their existing punctuation behavior; only digit characters change. The converter does not parse a number or change its value, date, or calendar. Recognized links and email addresses keep their original digits. Before an address cue appears, digits use the normal Nepali style; uninterrupted typing of the cue restores the current token to its original address spelling.

Existing Devanagari digits remain unchanged. Developers can choose ASCII digits with createEngine({ digits: 'latin' }), then pass that engine's convertWord and convertText functions to the browser adapter. This still converts Roman letters to Nepali. See digit configuration for app, page, and field setup. English mode keeps all input literal, including 0–9, and does not rewrite earlier text.

Nasal marks and punctuation

Type ^ after a syllable for bindu/anusvara ं, or ~ for chandrabindu ँ. These are explicit marks, so n and m still type consonants.

Type	Output
ka^	कं
kaa~	काँ

The period . always stays an English period, including in 3.14. Type | for Nepali पूर्णविराम ।: pani| → पनि। and 3.14| → ३.१४।. The same rule applies to convertText and text pasted into an attached field while Nepali mode is on, outside recognized technical spans. A pipe after a bare domain converts (camera.com| → camera.com।), but a pipe inside a URL path, query, or fragment stays literal (camera.com/a| stays as typed). Separate a URL suffix and sentence danda with whitespace: camera.com/a | → camera.com/a ।. The DOM controller also exposes insertPunctuation('।') and insertPunctuation('॥'); | types the single danda । only.

Alternatives and field editing

The engine returns candidates in preferred order. For example, kam gives कम first and काम second; cha gives च first and छ second. Use chha for the single छ reading. The field adapter shows the first reading inline and reports alternatives through onStateChange only while an ambiguous word is active. An integrating app can show a dropdown and call chooseCandidate(index); the adapter also handles Alt+1, Alt+2, and so on. Press Space to finish the word with the displayed reading. While a word is active, Backspace edits its original Roman sequence and recalculates the Nepali output.

The controller’s setEnabled(false) switches subsequent input to literal typing; setEnabled(true) resumes conversion. When conversion is disabled, ASCII digits and keys such as ^, ~, backtick, /, and | remain literal. Switching modes does not rewrite text already in the field. convertText always uses the first reading of each converted word and returns plain text without candidate data.

Mixed text and literal English

The default convertText and browser adapters preserve recognizable HTTP(S) links, www. links, ASCII domain-shaped hosts, and ordinary ASCII email addresses, including plus tags and subdomains. Their original spelling and case stay intact, while Nepali around them converts normally:

namaste camera.com name+tag@example.com
→ नमस्ते camera.com name+tag@example.com

namaste https://Example.com/a|b?q=camera#may pani|
→ नमस्ते https://Example.com/a|b?q=camera#may पनि।

Preservation starts at an early address cue: http:, https:, www., an ordinary ASCII local part followed by @, or the first letter after a domain dot. These incomplete examples already stay literal in live typing, paste, and whole-text conversion:

Type	Output
https:	https:
www.	www.
name@	name@
name@example	name@example
camera.c	camera.c

Before a cue appears, ordinary words still convert: camera → क्यामेरा and camera. → क्यामेरा. A trailing period alone does not select an address. During uninterrupted typing, the current token returns to Roman text as soon as the cue appears; it does not wait for a complete address. A space finishes the token. The adapter cannot recover Roman spellings for previously committed text. Switch to English mode before the first key for text that must stay literal from its beginning.

Address recognition is a pattern heuristic: no DNS or public-suffix check is performed, and preserved unfinished addresses are not validated. pani.paani and the unfinished pani.p stay literal because they look like domains. This policy has ASCII scope and does not provide full Unicode/internationalized address parsing or general English, code, or filename detection. convertWord remains a single-Roman-word converter; use convertText for mixed strings. A custom engine can disable preservation with createEngine({ preserveTechnicalText: false }).

For an English phrase or code fragment, switch the field’s controller to setEnabled(false) before typing or pasting it, then use setEnabled(true) to resume Nepali. A manager’s setEnabled() switches every field it owns. No existing text is rewritten. For fixed literal names, custom entries such as { github: ['GitHub'] } provide a configured spelling; for arbitrary English in bulk text, let the host app convert only its chosen Nepali chunks. See API examples.

Current limits

• The starter lexicon is small. Unknown words use deterministic phonetic rules, which can give incorrect Nepali spelling. There is no context-sensitive ranking or language detection.
• convertText preserves recognized technical spans, then converts Latin-letter runs outside them regardless of whether they are Nepali or English. The 51 loanword stems cover their listed keys and 19 finite attached suffix keys; the 12 month names keep exact lookup. These rules add no language detection or general inflection. Arbitrary English and code remain outside automatic preservation.
• The browser adapter attaches to <textarea> and text/search inputs, directly or through an opt-in field manager. It handles keyboard input, paste, and composition events, but this reference is not a browser compatibility guarantee.
• The prototype does not offer a dedicated keyboard shortcut for every Devanagari character, mark, or accent. A future language profile would need its own reviewed mappings; current behavior is Nepali-specific.

The implementation and regression examples are in phonetic rules, lexicon, engine, browser adapters, engine tests, and adapter tests. The Unicode Indic FAQ explains why character sequences and rendered Devanagari shapes must be considered separately.