Sinhala Phonetic Orthography

English

සිංහල

English

සිංහල

Appearance

03: Hal (al-lakuna ්) and consonant clusters / conjuncts ​

Scope: the Sinhala al-lakuna (hal kirīma), the three reduced forms (yansaya, rakāransaya, rēpaya), ligated conjuncts (bændi akuru), touching letters (sparśa akuru), cluster phonotactics, geminates, and the stored sequences that SLS 1134 and Unicode specify for them.

Compiled October 2026.

Conventions used in this file:

  • Code points are given as U+XXXX sequences. U+200D = ZERO WIDTH JOINER (ZWJ), U+200C = ZERO WIDTH NON-JOINER (ZWNJ), U+0DCA = SINHALA SIGN AL-LAKUNA (hal).
  • Every code point sequence in this file was machine-generated from the Sinhala string shown next to it, so the two always agree.
  • Letter IDs (used throughout this repository): ka kha ga … lla fa; vowels a aa … au; hal.
  • Confidence: high = stated directly by Unicode / SLS / font-shaping spec, or by two or more independent sources; medium = one good source, or a reasonable inference from good sources; low = not confirmed in a consulted source, or contested. [open] marks claims that no consulted source confirms; they are listed under Open questions. Counts such as "NLPC 108" are token counts in the NLPC 2.1M-word list (reports/corpus-counts.md); "rare" means fewer than 20 tokens.
  • Romanization in examples is informal (t = ට, th = ත, d = ඩ, dh = ද, sh = ශ, ss = ෂ, ə = schwa). It is not a proposal.

0. Sources (cited by ID below) ​

IDSourceType / authority
S1Unicode Standard 15.0, ch. 13 §13.2 "Sinhala": https://www.unicode.org/versions/Unicode15.0.0/ch13.pdfNormative-ish (core spec text)
S2Unicode NamedSequences.txt: https://www.unicode.org/Public/UCD/latest/ucd/NamedSequences.txtNormative data
S3SLS 1134:2004 draft for public comment (WG2 N2737 / L2/04-131), https://www.unicode.org/wg2/docs/n2737.pdf ; second copy https://sinhala.sourceforge.net/archive/akuru.org/att-0028/sls1134.pdfNational standard (draft). The final 2004 and the 2011 revision were not obtainable online.
S4SLSI working-group decisions 2004-06-09, L2/04-231: https://www.unicode.org/L2/L2004/04231-sinhala-rep.pdfStandards body minutes
S5Microsoft, "Creating and Supporting OpenType Fonts for Sinhala Script": https://learn.microsoft.com/en-us/typography/script-development/sinhalaShaping spec (pre-Win10 engine)
S6Microsoft, "Universal Shaping Engine" spec: https://learn.microsoft.com/en-us/typography/script-development/useShaping spec (current Windows engine for Sinhala)
S7HarfBuzz source: src/hb-ot-shaper.hh (Sinhala → USE) and src/hb-ot-shaper-use.cc (F_MANUAL_ZWJ), https://github.com/harfbuzz/harfbuzzImplementation
S8ICANN Sinhala Generation Panel, "Proposal for a Sinhala Script Root Zone LGR", 22 Apr 2019: https://new.icann.org/en/system/files/files/proposal-sinhala-lgr-22apr19-en.pdfExpert panel (UCSC, J.B. Disanayaka, et al.)
S9M. Jansche, unicode list, Oct 2016 (quotes SLS 1134:2011 Table 3 note), https://corp.unicode.org/pipermail/unicode/2016-October/004249.html ; reply (A. Freytag) https://unicode.org/mail-arch/unicode-ml/y2016-m10/0218.htmlSecondary quote of the standard
S10Harshula / R. Wordingham, unicode list, Oct 2018 "Fallback for Sinhala Consonant Clusters": https://www.unicode.org/mail-arch/unicode-ml/y2018-m10/0062.html ; https://www.unicode.org/mail-arch/unicode-ml/y2018-m10/0041.html ; https://corp.unicode.org/pipermail/unicode/2018-October/007087.htmlExpert discussion; reports SLS 1134:2011 content
S11Harshula, unicode list, Sep 2009 "Use of ZWJ to form Sinhala Conjuncts": https://unicode.org/mail-arch/unicode-ml/y2009-m09/0011.htmlExpert discussion
S12Harshula Jayasuriya, "Sinhala Orthography: Ola Leaf to the Computer", UCSC LTRL, 2007, https://ftp3.gwdg.de/pub/gnu/www/savannah-checkouts/non-gnu/sinhala/doc/presentations/sinhala-orthography-hj-20070212.pdfExpert presentation
S13Wasala, Weerasinghe, Gamage, "Sinhala Grapheme-to-Phoneme Conversion and Rules for Schwa Epenthesis", COLING/ACL 2006: https://aclanthology.org/P06-2114.pdfPeer-reviewed linguistics/NLP
S14Wikipedia, "Sinhala language" (Phonology): https://en.wikipedia.org/wiki/Sinhala_languageLead only
S15si.wikipedia "සිංහල හෝඩිය" conjunct table: https://si.wikipedia.org/?curid=2505Lead only
S16Usgoda Dhammagaru, "සිංහල භාෂාවේ අක්ෂර වින්‍යාසය" (teaching notes, 2024): https://ia601404.us.archive.org/21/items/Index_201704/dhgra002.pdfTeaching material (medium)
S17Noto Sans Sinhala sources (glyph inventory): https://github.com/notofonts/sinhalaFont implementation
S18fontconfig si.orth patch citing SLS 1134 Part 2:2007 compliance levels, https://bugs.freedesktop.org/attachment.cgi?id=22587Secondary quote of the standard
S19S. Wiles, Indology list, June 2012 "Sinhala ligatures": https://list.indology.info/pipermail/indology/2012-June/036715.htmlScholar's note
S20M. Kaplan, "The subtle difference between ශ්රී ලංකාව and ශ්‍රීලංකාව", 2007: https://archives.miloush.net/michkap/archive/2007/10/14/5448243.htmlImplementer blog
S21E. Muller (Adobe), comments on Sinhala draft, L2/04-235: https://www.unicode.org/L2/L2004/04235-sinhala-cmt.htmlUTC document
S22si.wikipedia "සිංහල අක්ෂර වින්‍යාසය": https://si.wikipedia.org/wiki/සිංහල_අක්ෂර_වින්‍යාසයLead only
S23A published Sinhala transliteration scheme (nongnu.org Sinhala project documentation): https://www.nongnu.org/sinhala/doc/transliteration/sinhala-transliteration_2.htmlPrior-art romanization
S24RFC 5892 (IDNA2008), Appendix A.2, CONTEXTJ rule for ZWJ: https://www.rfc-editor.org/rfc/rfc5892IETF standard (not consulted directly)
S25si.wikipedia articles "දුම්රිය" and "ශ්‍රී ලංකා දුම්රිය සේවය", wikitext fetched 2026-10-05: https://si.wikipedia.org/wiki/දුම්රියUsage sample (running text)

1. Things that will bite you (read first) ​

  1. In Sinhala, hal never joins anything by itself. C + ් + C always shows a visible hal. You only get a conjunct, yansaya, rakāransaya or rēpaya when ZWJ is present. Sinhala differs from Devanagari here. (HC-001)
  2. ZWJ order matters. ් + ZWJ gives a conjunct or reduced form. ZWJ + ් gives a touching letter. These are different spellings of the same sounds. (HC-002, HC-003)
  3. Yansaya and rakāransaya are mandatory in correct spelling. ක්රම (no ZWJ) is not an accepted spelling of krama. Rēpaya is optional. Other ligated conjuncts (ක්‍ෂ, න්‍ද …) are optional and mostly old-fashioned. (HC-020, HC-030, HC-040)
  4. ර + ් + ZWJ + ය is ambiguous (ra + yansaya, or rēpaya + ya). SLS 1134 says ra + yansaya is not used, so it means rēpaya + ya. Fonts disagree. (HC-033)
  5. ඥ is atomic (U+0DA5). Never build it as ජ්‍ඤ. (HC-051)
  6. A word that ends in a consonant sound is written with hal. The inherent vowel is pronounced as a or ə, but spelling never shows which. Schwa never triggers a hal. (HC-010…HC-013)
  7. Rakāransaya + u/ū have special glyph shapes that look like æ/ǣ signs. They must still be encoded with U+0DD4/U+0DD6. Real corpora contain the wrong encoding, e.g. ක්‍රෑර written for krūra. The usual spelling of krūra is in fact කෲර, with ෲ (02:VS-035). (HC-024)
  8. Prenasalized letters (ඟ ඦ ඬ ඳ ඹ) never take hal. ළ does not take hal either. (HC-014)
  9. Many systems strip or mangle ZWJ: some renderers, search boxes, the root-zone LGR. Expect ZWJ-less spellings in real-world corpora and fold them together when matching. (§10)

2. Encoding model ​

HC-001: Al-lakuna alone never forms a cluster ​

  • Statement: Without ZWJ, U+0DCA is always visible. It just "kills" the inherent vowel. It never triggers a conjunct, reduced form or touching form.
  • Examples: ද + ් + ධ → ද්ධ (visible hal on ද, then a full ධ): U+0DAF U+0DCA U+0DB0. Same with ධර්ම dharma written with explicit hal: U+0DB0 U+0DBB U+0DCA U+0DB8.
  • Exceptions: none. This is the base rule of Sinhala encoding. It differs from Devanagari/Tamil practice.
  • Applies-to: hal; all consonants.
  • Confidence: high.
  • Sources: S1 (§13.2 "Virama (al-lakuna) and Consonant Forms"), S12, S11.

HC-002: Three representations of a consonant pair ​

  • Statement: For a consonant pair C1 C2 (no vowel in between) there are three encodings. Each gives a different visual style:
    StyleSequenceUse
    Separate letters (modern)C1 ් C2default modern writing
    Conjunct / reduced formC1 ් ZWJ C2yansaya, rakāransaya, rēpaya, ligatures (bændi akuru)
    Touching lettersC1 ZWJ ් C2Pali and classical texts (sparśa akuru)
  • Examples (d + dh): separate ද්ධ U+0DAF U+0DCA U+0DB0; ligature ද්‍ධ U+0DAF U+0DCA U+200D U+0DB0; touching ද‍්ධ U+0DAF U+200D U+0DCA U+0DB0.
  • Exceptions: The ZWJ position does not decide which consonant takes the reduced form. That differs from Indic ZWJ use. The order only picks the style (S1).
  • Applies-to: hal + any consonant pair.
  • Confidence: high.
  • Sources: S1, S5 (cluster grammar {C+H+ZWJ}+C vs {C+ZWJ+H}+C), S12, S11, S10.

HC-003: Touching letters use ZWJ + al-lakuna (SLS 1134:2011, Unicode) ​

  • Statement: A touching cluster is C1 + ZWJ + ් + C2. Unicode says this style is productive. A font should handle it generically, not as a case-by-case list.
  • Examples: Pali kka ක‍්ක U+0D9A U+200D U+0DCA U+0D9A (S5's example); mma ම‍්ම U+0DB8 U+200D U+0DCA U+0DB8 (S15).
  • Exceptions / history: The 2004 draft SLS 1134 (S3) used ් + ZWJ for both conjuncts and touching letters. Its note on Pali says the sequence 0DCA 200D "may be" used. The SLSI 2004 minutes (S4) then asked Unicode what joiner to use. The candidates were ZWNJ, or ZWJ+ZWNJ+ZWJ. The final 2004 standard and the 2011 revision settled on ZWJ + al-lakuna, per Harshula (S10). Old documents may still use the draft convention.
  • Applies-to: all consonants; hal.
  • Confidence: high for current practice; medium for exactly when the change happened.
  • Sources: S1, S5, S10, S11, S12, S3, S4.

HC-004: Named sequences for the three reduced forms ​

  • Statement: Unicode defines named character sequences:
    • SINHALA CONSONANT SIGN YANSAYA = U+0DCA U+200D U+0DBA
    • SINHALA CONSONANT SIGN RAKAARAANSAYA = U+0DCA U+200D U+0DBB
    • SINHALA CONSONANT SIGN REPAYA = U+0DBB U+0DCA U+200D
  • Applies-to: ya, ra, hal.
  • Confidence: high.
  • Sources: S2, S1 (Table 13-3).

HC-005: ZWJ/ZWNJ semantics in Sinhala (summary) ​

  • Statement:
    • ZWJ after hal: request a conjunct or reduced form.
    • ZWJ before hal: request a touching form.
    • ZWNJ: no role in normal Sinhala text, because hal does not join by default. Its documented uses are marginal:
      • (a) The 2004 draft (S3) put ZWNJ before a vowel sign to show it standalone (U+200C U+0DCF). It also used ZWNJ for a standalone yansaya (U+200C U+0DCA U+200D U+0DBA) and a standalone rēpaya (U+0DBB U+0DCA U+200D U+200C).
      • (b) The SLSI 2004 decisions (S4) replaced these with SPACE-based sequences. Standalone yansaya = U+0020 U+0DCA U+200D U+0DBA. Standalone rakāransaya = U+0020 U+0DCA U+200D U+0DBB. Standalone rēpaya = U+0DBB U+0DCA U+200D U+0020. A standalone vowel sign = space + sign, without ZWJ.
      • (c) Jansche/Freytag (S9) proposed U+0DBB U+200C U+0DCA U+200D U+0DBA to force the non-standard "ra + yansaya" reading. This is a proposal only; no adoption is documented in the sources consulted.
    • Under USE (S6), ZWNJ blocks fusion of two characters. ZWJ joins the preceding cluster to the following character.
  • Applies-to: hal, ya, ra.
  • Confidence: high for ZWJ; medium for ZWNJ (the standalone conventions conflict between the draft and the WG decisions).
  • Sources: S1, S3, S4, S6, S9.

HC-006: Renderers: Sinhala is shaped by USE in Windows 10+ and HarfBuzz ​

  • Statement: Windows 10+ shapes Sinhala with the Universal Shaping Engine, which replaced the dedicated Sinhala engine. HarfBuzz also routes HB_SCRIPT_SINHALA to its USE shaper. Its USE basic features are applied with F_MANUAL_ZWJ, so ZWJ is not skipped and the font's lookups must match it explicitly. In practice:
    • Whether a conjunct appears depends entirely on the font having a lookup for C ් ZWJ C.
    • Fallback differs by platform when the font has no glyph. HarfBuzz shows a visible hal. Windows 10 and iOS also show hal but place vowels differently (S10).
  • Font coverage seen: Noto Sans Sinhala (S17) has ligature glyphs for exactly the Unicode Table 13-4 set plus ඤ්‍ච (glyph names ka_ssa, ka_va, ga_dha, tta_ttha, ta_tha, ta_va, da_dha, da_va, na_tha, na_da, na_dha, na_va, nya_ca). It also has a generic touch glyph and special rakāransaya forms for ද ඳ ඤ ඥ and the න්‍ද cluster.
  • Confidence: high (S6, S7 source code); medium for the platform fallback details (one mailing-list report).
  • Sources: S5, S6, S7, S10, S17.

3. Hal usage and pronunciation ​

HC-010: Hal marks the absence of a vowel; the inherent vowel is never written ​

  • Statement: A consonant letter with no sign carries the inherent vowel. A consonant with hal has no vowel. That is the only distinction spelling makes. Whether the inherent vowel is pronounced [a] or [ə] is not shown.
  • Examples: ගම gamə "village": U+0D9C U+0DB8. කර "shoulder" /karə/ vs කර "do" /kərə/ is a minimal pair spelled identically (S14).
  • Exceptions: none in spelling. S23, a transliteration scheme, maps both romanized a and e to the bare consonant, i.e. it lets ə be written as e.
  • Applies-to: all consonants; vowel a; hal.
  • Confidence: high.
  • Sources: S13, S14, S23.

HC-011: Schwa vs /a/ realisation rules (pronunciation only) ​

  • Statement: S13 starts by giving every bare consonant /ə/. It then applies ordered rules (paraphrased):

    1. The first syllable's nucleus becomes /a/. Exceptions: the word starts with /sv/; it starts with /kər/ (e.g. කර-, so කරනවා → /kərənəwaː/); or it is a single CV syllable.
    2. After a consonant + /r/, ə→a before a consonant; there are sub-rules for /h/.
    3. ə→a before /h/ when a vowel follows the /h/.
    4. ə→a before a consonant cluster (i.e. before a hal-consonant).
    5. ə→a before a word-final consonant, except final /r/, /b/, /ɖ/, /ʈ/.
    6. Word-final ə before /ji/ → a.
    7. /kə(r|l)u/ → /ka(r|l)u/.
    8. A small lexical rule for kal- forms.

    The rules reached 98% on 30,000 corpus words. Most remaining errors were compounds and English loans.

  • Examples: කරනවා: U+0D9A U+0DBB U+0DB1 U+0DC0 U+0DCF → /kərənəwaː/ by rule 1(b). Note: S13 gives a schwa in the first syllable too, not "karanəwa".

  • Implication: Hal placement is never driven by schwa. The romanizations karanawa, karanewa and kəranəwa all correspond to the same letters (ක ර න වා).

  • Exceptions: S13 found some words written without ා but pronounced with a long final vowel: අම්ම → /ammaː/, අක්ක, ගත්ත. Spelling is conservative here.

  • Applies-to: vowel a; all consonants.

  • Confidence: high (peer-reviewed); the rule summary is a paraphrase.

  • Sources: S13, S14.

HC-012: Word-final hal ​

  • Statement: A word ending in a consonant sound is written with a final hal. This is common in modern Sinhala. Typical sources:
    • inflectional endings: instrumental/ablative -in (මගින්, අතින්); indefinite -ak (පොතක්); -t "also" (මමත්); conditional -nam (නම්); plurals (මල්, ගස්, කන්)
    • native nouns (පින්, බත්)
    • loans (බස්, කැම්පස්)
  • Examples: මගින් U+0DB8 U+0D9C U+0DD2 U+0DB1 U+0DCA; පින් U+0DB4 U+0DD2 U+0DB1 U+0DCA; පොතක් U+0DB4 U+0DDC U+0DAD U+0D9A U+0DCA.
  • Exceptions / gotchas:
    • Word-final hal occurs only on letters that can take hal at all (see HC-014).
    • S14 says Sinhala words cannot phonetically end in a nasal other than /ŋ/. So final න් / ම් may be pronounced [ŋ] in speech, even though the spelling keeps න් / ම්. Medium confidence; only one lead source. The anusvara ං also writes /ŋ/ (e.g. සිංහල), so a romanized ng at word end is ambiguous between ං, න් and ම්.
  • Applies-to: hal; ka ta tha na ma la sa pa tta etc.
  • Confidence: high for the rule; medium for the [ŋ] note.
  • Sources: S13, S14, S8 (examples of hal use), general corpus observation.

HC-013: Mid-word hal = cluster or geminate ​

  • Statement: A hal inside a word always marks a consonant cluster or geminate with no vowel between (අම්මා, පත්තරය, කරන්න). If ZWJ is also present, the cluster is drawn as a conjunct or reduced form (HC-002).
  • Examples: අම්මා U+0D85 U+0DB8 U+0DCA U+0DB8 U+0DCF; කරන්න U+0D9A U+0DBB U+0DB1 U+0DCA U+0DB1.
  • Confidence: high.
  • Sources: S1, S3, S13.

HC-014: Letters that never take hal ​

  • Statement:
    • The five prenasalized letters (sannaka) ඟ ඦ ඬ ඳ ඹ cannot be followed by hal (S8, "constraint for Sannjakas"). They also cannot geminate (S14).
    • Retroflex ළ never takes hal (S22).
    • ඞ is the reverse case: it "is never combined with a vowel" and appears only in pure form (S3, Table 3 note 1, text partly garbled in the draft PDF; most likely means ඞ්).
  • Exceptions: si.wikipedia (S15) lists ඳ්‍ඨ, ඳ්‍ධ, ඳ්‍ව as conjuncts. That contradicts S8. S15 is treated as unreliable here (see Open questions).
  • Applies-to: nnga, nyja, nndda, nda, mba, lla (no hal); nga (hal only).
  • Confidence: medium-high (S8 is authoritative, S22 is a lead).
  • Sources: S8, S14, S22, S3.

4. Yansaya, rakāransaya, rēpaya ​

HC-020: Yansaya (්‍ය): ya after a pure consonant ​

  • Statement: When ය follows a pure consonant, it is written as the post-base yansaya. Encoding: C + ් + ZWJ + ය.
    • Required in normal text: SLS says yansaya and rakāransaya "are required in normal Sinhala text". S8 says forms like වාක්ය (without ZWJ) are not accepted.
    • Leave the ZWJ out only if, for some reason, the yansaya is deliberately not wanted (S3 note).
  • Examples: වාක්‍ය vākya U+0DC0 U+0DCF U+0D9A U+0DCA U+200D U+0DBA; සත්‍ය satya U+0DC3 U+0DAD U+0DCA U+200D U+0DBA; විද්‍යාව U+0DC0 U+0DD2 U+0DAF U+0DCA U+200D U+0DBA U+0DCF U+0DC0; අවශ්‍ය U+0D85 U+0DC0 U+0DC1 U+0DCA U+200D U+0DBA; රාජ්‍ය U+0DBB U+0DCF U+0DA2 U+0DCA U+200D U+0DBA; මනුෂ්‍ය U+0DB8 U+0DB1 U+0DD4 U+0DC2 U+0DCA U+200D U+0DBA; ද්‍රව්‍ය U+0DAF U+0DCA U+200D U+0DBB U+0DC0 U+0DCA U+200D U+0DBA; ශය්‍යා (ya+yansaya) U+0DC1 U+0DBA U+0DCA U+200D U+0DBA U+0DCF.
  • Vowels allowed on the yansaya (SLS Table 3): a, aa, u, uu, e, ee, o, oo (8 combinations shown, though the text says "7"), e.g. ක්‍යෝ U+0D9A U+0DCA U+200D U+0DBA U+0DDD. The vowel applies to the ය, not the first consonant. Other combinations are described as not used, though SLS does not forbid them (S3 §6.3).
  • Restriction: never after ර (HC-033).
  • Applies-to: all consonants except ra (and except the hal-less letters of HC-014) + ya.
  • Confidence: high.
  • Sources: S1, S2, S3, S5, S8, S9.

HC-021: Rakāransaya (්‍ර): ra after a pure consonant ​

  • Statement: When ර follows a pure consonant, it is written as the below-base rakāransaya. Encoding: C + ් + ZWJ + ර. It is mandatory in normal text: S8 says ක්රම is not accepted for ක්‍රම. Vowel signs follow the ර: C ් ZWJ ර V.
  • Examples:
    • ක්‍රම krama U+0D9A U+0DCA U+200D U+0DBB U+0DB8
    • ප්‍රශ්නය U+0DB4 U+0DCA U+200D U+0DBB U+0DC1 U+0DCA U+0DB1 U+0DBA
    • ප්‍රේමය U+0DB4 U+0DCA U+200D U+0DBB U+0DDA U+0DB8 U+0DBA
    • ශ්‍රී U+0DC1 U+0DCA U+200D U+0DBB U+0DD3
    • ග්‍රාමය U+0D9C U+0DCA U+200D U+0DBB U+0DCF U+0DB8 U+0DBA
    • ද්‍රව U+0DAF U+0DCA U+200D U+0DBB U+0DC0
    • බ්‍රාහ්මණ U+0DB6 U+0DCA U+200D U+0DBB U+0DCF U+0DC4 U+0DCA U+0DB8 U+0DAB
    • හ්‍රස්ව U+0DC4 U+0DCA U+200D U+0DBB U+0DC3 U+0DCA U+0DC0
    • ස්‍රාවය U+0DC3 U+0DCA U+200D U+0DBB U+0DCF U+0DC0 U+0DBA
    • ව්‍රත U+0DC0 U+0DCA U+200D U+0DBB U+0DAD
  • Vowels allowed on the rakāransaya (SLS Table 3): a, aa, ae, aee, i, ii, e, ee, ai, o, oo, au (12 combinations). Notice that u/uu are not in the table, but see HC-024.
  • Applies-to: effectively every consonant that can take hal + ra. Common with ka ga gha ta tha da dha pa ba bha va sha sa ha tta; rare with ma (HC-052); loans with fa dda.
  • Confidence: high.
  • Sources: S1, S2, S3, S5, S8.

HC-022: Rēpaya (ර්‍): ra before a consonant ​

  • Statement: When ර් comes before a consonant, it can be written as the above-base rēpaya. Encoding: ර + ් + ZWJ + C. Shapers reorder the glyph after the base consonant, and after the yansaya if there is one (S5).
    • The rēpaya is optional. Both කර්ම and කර්‍ම are valid (S3 §3.5). S8 likewise accepts both තර්ක and තර්‍ක.
  • Examples:
    • ධර්‍ම / ධර්ම dharma: U+0DB0 U+0DBB U+0DCA U+200D U+0DB8 vs U+0DB0 U+0DBB U+0DCA U+0DB8
    • තර්‍කය U+0DAD U+0DBB U+0DCA U+200D U+0D9A U+0DBA
    • මාර්‍ගය U+0DB8 U+0DCF U+0DBB U+0DCA U+200D U+0D9C U+0DBA
    • වර්‍ෂය U+0DC0 U+0DBB U+0DCA U+200D U+0DC2 U+0DBA
    • පූර්‍ව U+0DB4 U+0DD6 U+0DBB U+0DCA U+200D U+0DC0
    • අර්‍ථය U+0D85 U+0DBB U+0DCA U+200D U+0DAE U+0DBA
    • වර්‍ණ U+0DC0 U+0DBB U+0DCA U+200D U+0DAB
  • Applies-to: ra + hal + almost any consonant (ka ga ca ja nna ta tha da dha na pa ba bha ma ya va sha ssa sa ha).
  • Confidence: high (encoding); high (optionality).
  • Sources: S1, S2, S3, S5, S8.

HC-023: Stacking order and combinations ​

  • Statement: Logical order is always left to right in pronunciation order. ZWJ follows every hal that joins:
    • conjunct + rakāransaya: න්‍ද්‍රා U+0DB1 U+0DCA U+200D U+0DAF U+0DCA U+200D U+0DBB U+0DCF (SLS 1134 §5.8 example)
    • consonant + yansaya + rakāransaya (si.wikipedia pattern): ක්‍ය්‍ර U+0D9A U+0DCA U+200D U+0DBA U+0DCA U+200D U+0DBB (rare; pattern only)
    • rēpaya + consonant + yansaya: ර්‍ක්‍ය U+0DBB U+0DCA U+200D U+0D9A U+0DCA U+200D U+0DBA (pattern only)
    • vowel signs come last: ක්‍රෝ U+0D9A U+0DCA U+200D U+0DBB U+0DDD, ප්‍රේ U+0DB4 U+0DCA U+200D U+0DBB U+0DDA
    • The kombuva (ෙ) is a pre-base glyph. It is still encoded after the whole cluster.
  • Confidence: high.
  • Sources: S3 §5.6–5.8, S5, S15.

HC-024: Rakāransaya + u / uu: special shapes, normal encoding ​

  • Statement: After a rakāransaya, the u and uu vowel signs take alternative shapes that look like the æ/ǣ signs. Unicode says they must be encoded as U+0DD4/U+0DD6, not as U+0DD0/U+0DD1.
  • Examples: ක්‍රු U+0D9A U+0DCA U+200D U+0DBB U+0DD4, ක්‍රූර krūra U+0D9A U+0DCA U+200D U+0DBB U+0DD6 U+0DBB. Wrong (but attested): ක්‍රෑර U+0D9A U+0DCA U+200D U+0DBB U+0DD1 U+0DBB. S13 found corpus words where ැ/ෑ were used for /u, uː/. S13 printed its examples in a legacy font, so they are garbled. They are most likely ශ්‍රැති-type and ක්‍රෑර-type spellings.
  • Also: Bases shaped like ද lose their tail before a below-base sign. ද්‍ර looks different from ක්‍ර, but the encoding is unaffected (S1).
  • Usage and rendering (02:VS-035): in practice /Cru/ and /Cruː/ are mostly written C + ෘ/ෲ (කෲර 1,667 vs ක්‍රූර 9 vs ක්‍රෑර 186 in the NLPC 2.1M-word list, CC-03). Several fonts, including the Windows system fonts Nirmala UI and Iskoola Pota, do not draw the special shape and attach an ordinary ු/ූ below the cluster.
  • Applies-to: ra (rakāransaya), u, uu, ae, aee.
  • Confidence: high (S1); medium for the corpus interpretation.
  • Sources: S1, S13, S3 §6.3 note.

5. Ligated conjuncts (bændi akuru) and touching letters (sparśa akuru) ​

HC-030: Ligated conjuncts are optional and mostly classical ​

  • Statement: Apart from yansaya and rakāransaya, Sinhala has no strict obligatory ligatures in modern writing. Harshula describes modern orthography as "separate letters" (S12). Wiles calls bændi akuru "optional", with few in common modern use (S19). Encoding: C1 ් ZWJ C2.
  • Classification (S8):
    • Conjuncts still met today: ක්‍ෂ, ක්‍ව, න්‍ද, න්‍ධ, න්‍ථ, ත්‍ථ (the S8 PDF lost its glyphs; the names are given as kSa, kva, nda, ndha, ntha, ttha).
    • Not used in contemporary writing: ද්‍ධ, ද්‍ව, ට්‍ඨ, ඤ්‍ච.
  • Examples: අක්‍ෂරය U+0D85 U+0D9A U+0DCA U+200D U+0DC2 U+0DBB U+0DBA vs අක්ෂරය U+0D85 U+0D9A U+0DCA U+0DC2 U+0DBB U+0DBA; බුද්‍ධ U+0DB6 U+0DD4 U+0DAF U+0DCA U+200D U+0DB0 vs බුද්ධ U+0DB6 U+0DD4 U+0DAF U+0DCA U+0DB0; ආනන්‍ද U+0D86 U+0DB1 U+0DB1 U+0DCA U+200D U+0DAF.
  • Confidence: high that they are optional; medium for the exact "still used" list.
  • Sources: S1 (Table 13-4), S8, S12, S19, S17.

HC-031: Touching letters are for Pali/classical text only ​

  • Statement: Touching letters show a pure consonant drawn against the next letter instead of with a hal. Encoding: C1 ZWJ ් C2. They are used in old Sinhala and are frequent in Pali, but not in contemporary Sinhala (S8, S5, S19).
    • Typical pairs (S8): kka, kkha, gga, ccha, jja, jjha, ṭṭha, ppha, mma, and others.
    • SLS 1134 Part 2 makes touching letters required only at font compliance Level 3. As of 2009 no Level 3 font was known (S11, S18).
  • Examples: ධම‍්ම dhamma U+0DB0 U+0DB8 U+200D U+0DCA U+0DB8; බුද‍්ධ U+0DB6 U+0DD4 U+0DAF U+200D U+0DCA U+0DB0.
  • Confidence: high.
  • Sources: S1, S5, S8, S11, S18, S19.

HC-032: ක්ෂ vs ක්‍ෂ ​

  • Statement: Both are valid encodings of the same cluster kṣ:

    • ක්ෂ (hal visible) U+0D9A U+0DCA U+0DC2 is the modern "separate letters" spelling;
    • ක්‍ෂ U+0D9A U+0DCA U+200D U+0DC2 is the traditional ligature.

    The S16 teaching note mixes the two: රක්ෂණය next to නිරීක්‍ෂණ, අක්‍ෂර. Sinhala text in the wild uses both. A search or spell-check must treat them as equivalent.

  • Spelling rule (not encoding): kṣ is spelled with ෂ (ssa), not ශ or ස, e.g. අක්‍ෂර, ලක්‍ෂ (S22, S16).

  • Confidence: high.

  • Sources: S1, S8, S16, S22.

HC-033: ර ් ZWJ ය: rēpaya + ya, never ra + yansaya ​

  • Statement: SLS 1134 (Table 3 note, in both the 2004 draft and the 2011 revision as quoted by S9) says yansaya is not used after ර. It gives a ra+yansaya spelling of kārya as an example of incorrect spelling. So ර ් ZWJ ය means rēpaya over ය.
  • Accepted spellings of kārya (S8):
    • the traditional කාර්‍ය්‍ය: an extra ය is inserted, carrying both the rēpaya and a yansaya. U+0D9A U+0DCF U+0DBB U+0DCA U+200D U+0DBA U+0DCA U+200D U+0DBA. This is also the SLSI 2004 decision #3 (S4) and Microsoft's abvs example ර්‍ය්‍ය (S5).
    • the contemporary කාර්‍ය (rēpaya + ya, no yansaya), "also accepted" U+0D9A U+0DCF U+0DBB U+0DCA U+200D U+0DBA
    • කාර්ය (no rēpaya, no yansaya) U+0D9A U+0DCF U+0DBB U+0DCA U+0DBA, used by "those who do not follow the above writing system"
  • Conflict: The 2004 draft (S3 §5.7) encoded "yansaya with repaya" as U+0DBB U+0DCA U+200D U+200C U+0DCA U+200D U+0DBA (with ZWNJ). The SLSI decisions (S4) replaced this with U+0DBB U+0DCA U+200D U+0DBA U+0DCA U+200D U+0DBA. Fonts vary in how they render ර ් ZWJ ය (S9).
  • Same family: සූර්‍ය (NLPC 108), ආර්‍ය (179), ධෛර්‍ය (92), ආශ්චර්‍ය (62), ආචාර්‍ය (404). Their older spellings take the doubled ය්‍ය form. That form is attested for each word but is uncommon: counting ර්ය්‍ය with and without ZWJ after ර්, NLPC has සූර්ය්‍ය 21, ආර්ය්‍ය 81, ධෛර්ය්‍ය 41, ආශ්චර්ය්‍ය 18 (rare) and ආචාර්ය්‍ය 53 tokens. Plain-hal spellings are far more frequent (සූර්ය 3,193, ආචාර්ය 19,206; compare CC-07, CC-08).
  • Applies-to: ra, ya, hal.
  • Confidence: high on the rule; medium on which modern variant is preferred.
  • Sources: S3, S4, S5, S8, S9.

HC-034: ර ් ZWJ ර ​

  • Statement: The same ambiguity exists for ra + rakāransaya vs rēpaya + ra (S9). It is very rare in real words [no attested word found]. Treat it as rēpaya + ra by analogy with HC-033.
  • Confidence: low.
  • Sources: S9.

6. Clusters: native vs loan, geminates, triples ​

HC-040: Native Sinhala syllables are (C)V(C); clusters come from Sanskrit/Pali/English ​

  • Statement: Native words are limited to (C)V(C), V̄ and CV̄(C) syllables, with only marginal CC (S14). Native words have:

    • no initial clusters;
    • medial clusters only as coda + onset across a syllable boundary (mostly geminates and nasal + stop).

    Initial clusters (ප්‍ර-, ක්‍ර-, ශ්‍ර-, ස්ව-, ද්ව-, ත්‍ර-, ස්ථ-) belong to tatsama (Sanskrit/Pali) words and English loans (e.g. ස්ටේෂන් "station"). Uneducated speech tends to break them up (S14).

  • Examples: ප්‍රශ්නය, ක්‍රමය, ශ්‍රී, ස්වභාවය U+0DC3 U+0DCA U+0DC0 U+0DB7 U+0DCF U+0DC0 U+0DBA, ද්වාරය U+0DAF U+0DCA U+0DC0 U+0DCF U+0DBB U+0DBA, ස්ථානය U+0DC3 U+0DCA U+0DAE U+0DCF U+0DB1 U+0DBA.

  • Confidence: high (structure); medium (exact list).

  • Sources: S14, S1, S8.

HC-041: Initial clusters and the ZWJ they need ​

  • Statement: For word-initial clusters:
    • with ර or ය as second member: always use the reduced form, i.e. ZWJ is required (HC-020/021);
    • with any other second member (ස්ව, ද්ව, ස්ථ, ස්ක, ස්ප, ශ්ව …): explicit hal, no ZWJ;
    • with ව after ක/ත/ද/න: a ligature is possible (ක්‍ව, ත්‍ව, ද්‍ව, න්‍ව are in Unicode Table 13-4) but optional.
  • Examples: ස්ව U+0DC3 U+0DCA U+0DC0 (no ZWJ); ද්වාරය vs classical ද්‍වාරය U+0DAF U+0DCA U+200D U+0DC0 U+0DCF U+0DBB U+0DBA.
  • Confidence: high.
  • Sources: S1, S3, S8.

HC-042: Geminates are written C ් C (no ZWJ) ​

  • Statement: Doubled consonants are written with an explicit hal on the first copy. They are very common in native words: අම්මා, අක්කා, පත්තරය, බල්ලා, කරන්න. Use touching (ZWJ ්) only for Pali.
    • Not all letters geminate. The exceptions are the prenasalized letters, ඞ (/ŋ/), ෆ, හ and ශ (S14).
    • Where morphology would geminate හ, the result is ස්ස (S14).
    • Prenasalized letters revert to nasal + stop: ඳ → න්ද.
  • Examples: අම්මා U+0D85 U+0DB8 U+0DCA U+0DB8 U+0DCF; අක්කා U+0D85 U+0D9A U+0DCA U+0D9A U+0DCF; පත්තරය U+0DB4 U+0DAD U+0DCA U+0DAD U+0DBB U+0DBA; බල්ලා U+0DB6 U+0DBD U+0DCA U+0DBD U+0DCF; කණ්ණාඩිය U+0D9A U+0DAB U+0DCA U+0DAB U+0DCF U+0DA9 U+0DD2 U+0DBA.
  • Applies-to: ka ga ca ja tta dda nna ta da na pa ba ma ya la va sa. Not lla: ළ්ළ is not used (HC-014).
  • Confidence: high.
  • Sources: S14, S13, S8.

HC-043: තත්ත්වය vs තත්වය (and සත්ත්ව vs සත්ව) ​

  • Statement: The Sanskrit abstract suffix -tva is added to stems ending in -t (tat, sat). The prescriptive rule (S16) keeps both t's in tatsama words: තත්ත්වය U+0DAD U+0DAD U+0DCA U+0DAD U+0DCA U+0DC0 U+0DBA, සත්ත්වයා U+0DC3 U+0DAD U+0DCA U+0DAD U+0DCA U+0DC0 U+0DBA U+0DCF, තත්ත්වඥ. The reduced spellings තත්වය U+0DAD U+0DAD U+0DCA U+0DC0 U+0DBA and සත්වයා are the modern forms and are accepted as valid. Both spellings are frequent: තත්ත්වය NLPC 25,852 vs තත්වය 21,484, සත්ත්වයා 539 vs සත්වයා 1,451. The double-t form remains the grammatical (etymological) one; neither is a misspelling.
  • Note: The ත්ව part may also be drawn as the ligature ත්‍ව U+0DAD U+0DCA U+200D U+0DC0 (Unicode Table 13-4). That gives a third and fourth spelling (තත්ත්‍වය). Treat all as equivalent for search.
  • Confidence: medium. S16 is a teaching note, not an official decree. No Official Languages Department or NIE ruling was found.
  • Sources: S16; Wiktionary lists Sanskrit तत्त्व as තත්ත්ව (lead only).

HC-044: Triple (and longer) clusters ​

  • Statement: These occur only in tatsama words. Each joining point is encoded independently: ZWJ only where a reduced form or ligature is wanted, plain hal elsewhere.
  • Examples:
    • ස්ත්‍රී strī U+0DC3 U+0DCA U+0DAD U+0DCA U+200D U+0DBB U+0DD3 (s-hal, t + rakāransaya)
    • රාෂ්ට්‍ර rāṣṭra U+0DBB U+0DCF U+0DC2 U+0DCA U+0DA7 U+0DCA U+200D U+0DBB
    • ක්‍ෂේත්‍ර kṣetra: ligature form U+0D9A U+0DCA U+200D U+0DC2 U+0DDA U+0DAD U+0DCA U+200D U+0DBB; hal form U+0D9A U+0DCA U+0DC2 U+0DDA U+0DAD U+0DCA U+200D U+0DBB
    • ශාස්ත්‍ර U+0DC1 U+0DCF U+0DC3 U+0DCA U+0DAD U+0DCA U+200D U+0DBB
    • ඉන්ද්‍රිය U+0D89 U+0DB1 U+0DCA U+0DAF U+0DCA U+200D U+0DBB U+0DD2 U+0DBA
    • මන්ත්‍ර U+0DB8 U+0DB1 U+0DCA U+0DAD U+0DCA U+200D U+0DBB
    • ලක්‍ෂ්‍ය lakṣya U+0DBD U+0D9A U+0DCA U+200D U+0DC2 U+0DCA U+200D U+0DBA
    • මත්ස්‍ය matsya U+0DB8 U+0DAD U+0DCA U+0DC3 U+0DCA U+200D U+0DBA
    • චන්ද්‍ර U+0DA0 U+0DB1 U+0DCA U+0DAF U+0DCA U+200D U+0DBB
  • Confidence: high for the encoding; high for the words (common vocabulary).
  • Sources: S1, S3 §5.8.

7. Special cases ​

HC-050: ශ්‍රී ​

  • Statement: śrī is ශ + hal + ZWJ + ර + ී: U+0DC1 U+0DCA U+200D U+0DBB U+0DD3. Without ZWJ, ශ්රී U+0DC1 U+0DCA U+0DBB U+0DD3 shows a visible hal. That is "not conventional", but programmers often accept it (S20). Both are in wide circulation, so search must fold them together.
  • Confidence: high.
  • Sources: S20, S1.

HC-051: ජ්ඤ / ඥ ​

  • Statement: The conjunct j + ñ is atomically encoded as ඥ U+0DA5 (S1). It should not be built from ජ ් ZWJ ඤ.
    • Pronunciation: ඤ and ඥ sound the same only word-initially (ඤාණ / ඥාන). Elsewhere ඥ behaves as two consonant sounds, e.g. ප්‍රඥා (S3 §3.2 note 2).
    • Script status: S8 says the ඥ glyph is considered to represent j + ñ, and that it is a regular letter of contemporary Sinhala.
  • Examples: ඥානය U+0DA5 U+0DCF U+0DB1 U+0DBA, ප්‍රඥාව U+0DB4 U+0DCA U+200D U+0DBB U+0DA5 U+0DCF U+0DC0, විඥාන U+0DC0 U+0DD2 U+0DA5 U+0DCF U+0DB1.
  • Applies-to: jnya, ja, nya.
  • Confidence: high.
  • Sources: S1, S3, S8.

HC-052: Rakāransaya after ම / න / ල ​

  • Statement: Nothing in Unicode, SLS or the shaping specs restricts which consonant may take a rakāransaya. Any C ් ZWJ ර is valid encoding, and fonts draw the generic below-base form.
    • ම්‍ර in Sanskrit tatsama words is rare in running text: තාම්‍ර "copper" (rare: NLPC 3) and ආම්‍ර "mango" (rare: NLPC 4, with endings only). The plain spellings තාම්ර and ආම්ර do not occur as bare words either.
    • න්‍ර and ල්‍ර: CC-10 counts 269 and 100 tokens, against 10,133 and 2,786 with plain hal. The most frequent joined forms are ZWJ spellings of words normally written with plain hal (හෙන්‍රි 36 vs හෙන්රි 3,058). Every other න්‍ර or ල්‍ර form has fewer than 20 tokens.
    • In running text, ම න ල + ර is written with plain hal. Where ම න ල meets ර there is usually a morpheme or syllable boundary: දුම් + රිය "train", හෙන්රි "Henry", ඉම්රාන්, දිල්රුක්ෂි. In the NLPC 2.1M-word list, ම්ර outnumbers ම්‍ර 46,496 to 383 tokens and න්ර outnumbers න්‍ර 10,133 to 269 (දුම්රිය 30,800, දුම්‍රිය 82), while ZWJ is written in 98% of other C + ර tokens (CC-09, C-10). Sinhala Wikipedia agrees: 1,148 articles have ම්ර and 63 have ම්‍ර (CC-16). A ZWJ would draw a rakāransaya under ම, a form readers do not use for these words.
    • So the general rule of HC-021 (C + ර takes ZWJ) holds within a syllable. After ම න ල the plain form is the norm, and ම්‍ර is kept for a deliberately classical style.
  • Confidence: medium (encoding high; usage from one source).
  • Sources: S1, S3, S5, S17, S25.

HC-053: ඤ්‍ච / ඤ්‍ජ (pañca, vyañjana) ​

  • Statement:
    • Noto has a ligature glyph for ඤ්‍ච (nya_ca), and S8 lists "njca" among obsolete conjuncts. Modern spelling uses hal: පඤ්ච U+0DB4 U+0DA4 U+0DCA U+0DA0, ව්‍යඤ්ජන U+0DC0 U+0DCA U+200D U+0DBA U+0DA4 U+0DCA U+0DA2 U+0DB1.
    • A dedicated ligature for ඤ්‍ජ is not documented in the sources consulted [open].
    • Do not confuse either with ඦ (nyja), the archaic prenasalized letter that never takes hal.
  • Confidence: medium.
  • Sources: S8, S17, S3.

HC-054: ර් before ය / ර inside words ​

  • Statement: See HC-033 and HC-034. When a romanized r + y follows a vowel, there are three possible Sinhala spellings:

    • rēpaya + ya (ර ් ZWJ ය);
    • explicit hal (ර ් ය);
    • the traditional doubled form (ර ් ZWJ ය ් ZWJ ය).

    ra + yansaya is never a valid choice.

  • Confidence: high.

  • Sources: S3, S8, S9.

8. What SLS 1134 specifies for stored text ​

HC-060: Internal representation (stored sequences) ​

  • Statement: SLS 1134 (§5) fixes the stored sequences, and conforming text must use them:
    • pure consonant = C ්;
    • rakāransaya/yansaya = C ් ZWJ ර/ය;
    • rēpaya = ර ් ZWJ C;
    • other conjuncts = C ් ZWJ C;
    • touching = C ZWJ ් C (2011);
    • composite vowel signs: use the single precomposed sign. For example, කෝ is ක + ෝ (U+0DDD), not ක + ෙ + ා + ්. Sequences of separate signs are "discouraged" (S3 §5.4 note 2).
  • Confidence: high.
  • Sources: S3, S1, S10.

HC-062: Font compliance levels (SLS 1134 Part 2:2007) ​

  • Statement: There are three font levels.

    • Level 1 = full repertoire minus ඏ ඐ ෟ ෳ ෴.
    • Touching letters are required only at Level 3.

    Text containing touching sequences will therefore look broken on most systems.

  • Confidence: medium (secondary quotes).

  • Sources: S18, S11.

9. Complete table of known conjunct / reduced / touching forms ​

Legend:

  • Status: M = mandatory in correct modern spelling; O = optional (both forms accepted); C = classical/Pali only (rare today); P = productive pattern (any consonant).
  • Src: U = Unicode Table 13-3/13-4; N = present in Noto Sans Sinhala; I = ICANN LGR; W = si.wikipedia; S = SLS 1134; MS = Microsoft spec.
  • Letter IDs are those used throughout this repository.

9a. Reduced forms (productive) ​

FormLettersStatusSequenceExampleSrc
ක්‍ය (yansaya)C + hal + yaMU+0D9A U+0DCA U+200D U+0DBAවාක්‍යU S MS I
ක්‍ර (rakāransaya)C + hal + raMU+0D9A U+0DCA U+200D U+0DBBක්‍රමයU S MS I
ර්‍ක (rēpaya)ra + hal + COU+0DBB U+0DCA U+200D U+0D9Aතර්‍කයU S MS I
ර්‍ය්‍ය (rēpaya+ya+yansaya)ra ya yaO (traditional)U+0DBB U+0DCA U+200D U+0DBA U+0DCA U+200D U+0DBAකාර්‍ය්‍යS(2004 WG) MS I
conjunct + rakāransayae.g. na da raM for the -rU+0DB1 U+0DCA U+200D U+0DAF U+0DCA U+200D U+0DBBචන්ද්‍ර (lig.)S
ස්ත්‍ර (hal + rakāransaya)sa ta raM for the -rU+0DC3 U+0DCA U+0DAD U+0DCA U+200D U+0DBBස්ත්‍රී-

9b. Ligated conjuncts (C1 ් ZWJ C2): all optional ​

FormIDsStatusSequenceExample wordSrc
ක්‍ෂka+ssaO (still common)U+0D9A U+0DCA U+200D U+0DC2අක්‍ෂරය, ලක්‍ෂU N I W
ක්‍වka+vaOU+0D9A U+0DCA U+200D U+0DC0පක්‍වU N I W MS
ග්‍ධga+dhaO / CU+0D9C U+0DCA U+200D U+0DB0දුග්‍ධ, මුග්‍ධU N
ට්‍ඨtta+tthaC (not contemporary per I)U+0DA7 U+0DCA U+200D U+0DA8අට්‍ඨකථා (Pali)U N I
ත්‍ථta+thaO / CU+0DAD U+0DCA U+200D U+0DAE(Pali) අත්‍ථU N I W
ත්‍වta+vaOU+0DAD U+0DCA U+200D U+0DC0සත්‍ව, මහත්‍වයU N W
ද්‍ධda+dhaC (not contemporary per I)U+0DAF U+0DCA U+200D U+0DB0බුද්‍ධU N I
ද්‍වda+vaC (not contemporary per I)U+0DAF U+0DCA U+200D U+0DC0ද්‍විත්වU N I
න්‍ථna+thaOU+0DB1 U+0DCA U+200D U+0DAEග්‍රන්‍ථU N I
න්‍දna+daOU+0DB1 U+0DCA U+200D U+0DAFආනන්‍ද, චන්‍ද්‍රU N I W
න්‍ධna+dhaOU+0DB1 U+0DCA U+200D U+0DB0සන්‍ධි, බන්‍ධනU N I W
න්‍වna+vaOU+0DB1 U+0DCA U+200D U+0DC0අන්‍වයU N
ඤ්‍චnya+caC (not contemporary per I)U+0DA4 U+0DCA U+200D U+0DA0පඤ්‍චN I
ඥja+nyaM (atomic letter U+0DA5)U+0DA5ඥානයU I
ඳ්‍ඨ, ඳ්‍ධ, ඳ්‍වnda+…doubtfulU+0DB3 U+0DCA U+200D U+0DB0-W only (conflicts with I)

9c. Touching letters (C1 ZWJ ් C2): Pali/classical, productive ​

FormIDsSequenceNoteSrc
ක‍්කka+kaU+0D9A U+200D U+0DCA U+0D9AMS exampleMS I
ක‍්ඛka+khaU+0D9A U+200D U+0DCA U+0D9BI
ග‍්ගga+gaU+0D9C U+200D U+0DCA U+0D9CI
ච‍්ඡca+chaU+0DA0 U+200D U+0DCA U+0DA1I
ජ‍්ජja+jaU+0DA2 U+200D U+0DCA U+0DA2I
ජ‍්ඣja+jhaU+0DA2 U+200D U+0DCA U+0DA3I
ට‍්ඨtta+tthaU+0DA7 U+200D U+0DCA U+0DA8I
ප‍්ඵpa+phaU+0DB4 U+200D U+0DCA U+0DB5I
ම‍්මma+maU+0DB8 U+200D U+0DCA U+0DB8dhammaW I
ද‍්ධda+dhaU+0DAF U+200D U+0DCA U+0DB0Unicode exampleU
any C1 + C2-C1 U+200D U+0DCA C2Unicode: "productive"U

10. Implications for romanization and transliteration ​

These points apply to anyone converting between a Latin-script romanization and Sinhala script, or validating and searching stored Sinhala text.

  1. Every vowel-less consonant carries hal. A romanized consonant followed by another consonant or a word boundary corresponds to C + ්. Hal is never inferred from schwa: a and ə/e-for-schwa both mean "no sign" (HC-010, HC-011).
  2. C + r and C + y take ZWJ. C + r / C + y correspond to C ් ZWJ ර/ය (rakāransaya/yansaya). It is the only accepted modern spelling (HC-020, HC-021). Exception: after ර, ry never means ra + yansaya (HC-033).
  3. Rēpaya is a style choice, not grammar. rC can correspond to ර් C (hal) or ර ් ZWJ C (rēpaya); both are correct. A converter should document which it produces; rēpaya matches the traditional printed form. ර්ය also has a plain-hal option (HC-022, HC-033).
  4. Ligated conjuncts are optional. Whether kṣ is rendered ක්ෂ or ක්‍ෂ is a style decision, because both are common (HC-032). All other bændi akuru (ද්‍ධ, න්‍ද, ත්‍ව …) are classical: plain hal is the modern spelling, with ZWJ only for a deliberately classical style (HC-030).
  5. Touching letters belong to Pali/classical text. They are encoded C ZWJ ්, and most fonts below SLS Level 3 render them poorly (HC-031, HC-062).
  6. ඥ needs its own romanization (e.g. GN/jny). It is never composed as ජ ් ZWJ ඤ (HC-051).
  7. Geminates: a doubled consonant (amma, akka, paththaraya) → C ් C, no ZWJ (HC-042). For -tva words both තත්ත්වය and තත්වය are accepted: තත්ත්වය is the grammatical form, තත්වය the modern one (HC-043).
  8. Prenasalized letters and ළ cannot take hal. Romanized nd + consonant resolves to න්ද… (dental n + hal), not ඳ් (HC-014).
  9. Word-final nasal ambiguity: final ng could be ං, න් or ම් (HC-012). Resolving it needs a dictionary, not a fixed rule.
  10. Vowel on clusters goes on the last consonant. kre → ක්‍රෙ (ක ් ZWJ ර ෙ). kyoo → ක්‍යෝ. Use the precomposed vowel sign (ෝ U+0DDD etc.), never a split sequence (HC-060).
  11. kru/kruu written with rakāransaya use ු/ූ (U+0DD4/U+0DD6), even though the glyph looks like ැ/ෑ (HC-024). The more common spelling is C + ෘ/ෲ (කෘ, කෲ; 02:VS-035).
  12. Normalisation/search: fold C ් ZWJ C, C ් C and C ZWJ ් C together for matching. Writers and data sources mix them (ශ්‍රී/ශ්රී, අක්‍ෂර/අක්ෂර). Never strip ZWJ from stored or converted text. IDNA permits ZWJ after a virama (S24), but the root-zone LGR excludes it (S8).
  13. ZWNJ does not belong in normal text (HC-005).
  14. Writing order vs storage order. A romanization is naturally in logical order. For rēpaya, r comes first and the stored form is ර ් ZWJ before the base consonant, which is already logical order, even though the rēpaya is drawn above the following consonant (HC-022).

11. Open questions / conflicting sources ​

  1. Touching-letter encoding history. The 2004 draft SLS used ් ZWJ for both conjuncts and touching letters. The SLSI June 2004 minutes left the joiner undecided (ZWNJ or ZWJ+ZWNJ+ZWJ). Unicode, Microsoft and (per Harshula) SLS 1134:2011 use ZWJ ්. The final SLS 1134:2004 and 2011 texts were not available to verify first-hand. (S3, S4, S10)

  2. Yansaya-with-rēpaya encoding. Three candidates:

    • draft SLS: ර ් ZWJ ZWNJ ් ZWJ ය
    • SLSI WG 2004: ර ් ZWJ ය ් ZWJ ය
    • modern simplified: ර ් ZWJ ය (rēpaya on ya)

    Fonts render ර ් ZWJ ය inconsistently (S9). Which does SLS 1134:2011 mandate? Open: the 2011 text was not consulted directly.

  3. Standalone signs: ZWNJ-based (2004 draft) vs SPACE-based (SLSI WG 2004 decision). Irrelevant for running text, but matters for documentation that shows a sign on its own.

  4. ඳ + hal conjuncts (ඳ්‍ධ etc.) appear in si.wikipedia (S15). ICANN (S8) says sannaka letters never take hal. S15 is likely wrong or archaic.

  5. Which ligatures count as "still used". ICANN puts ක්‍ෂ, ක්‍ව, න්‍ද, න්‍ධ, න්‍ථ, ත්‍ථ in current use, but its text is garbled in the PDF. Harshula says modern writing has no strict ligation except r/y forms. No corpus frequency data for these ligatures has been compiled. Whether ඤ්‍ජ has a dedicated ligature (HC-053) is also open.

  6. ක්ෂ vs ක්‍ෂ preference in current NIE textbooks: unknown. The S16 notes use both.

  7. තත්ත්වය vs තත්වය: Both are accepted (HC-043). Only a teaching note (S16) states the grammatical double-t form in writing, and no official ruling (Official Languages Dept / NIE "Sinhala Lekhana Rithiya", 1989) was available for consultation on whether the modern form is formally sanctioned.

  8. Word-final න්/ම් → [ŋ] (S14) is one lead source. Not checked against Gair/Karunatillake.

  9. ම්‍ර, න්‍ර, ල්‍ර (resolved). The corpus count (CC-10) gives ම්ර 46,496 vs ම්‍ර 383, න්ර 10,133 vs න්‍ර 269 and ල්ර 2,786 vs ල්‍ර 100 tokens. The joined forms are mostly ZWJ spellings of plain-hal words (දුම්‍රිය 82, හෙන්‍රි 36), and tatsama ම්‍ර words are rare (තාම්‍ර 3). See HC-052.

  10. SLS Table 3 count: it says "7" yansaya combinations but lists 8. It also omits u/uu with rakāransaya, while Unicode shows ක්‍රු/ක්‍රූ shapes. This may be an editorial slip in the draft.

  11. "Sinhala Lekhana Rithiya" (NIE 1989) is cited by S8 as the alphabet source. It was not accessed and is likely the best source for school-level hal/conjunct rules.

  12. Corpus evidence of wrong encodings (ැ/ෑ for u/uu after rakāransaya). S13's examples are in a garbled legacy font, so the exact words are inferred.

  13. Removed rules: HC-025, HC-061, removed: out of scope.