noomorph · Aug 20, 2026 · Aug 20, 2026 · Aug 20, 2026 · Aug 20, 2026 · Aug 20, 2026
diff --git a/docs/conllu-conventions.md b/docs/conllu-conventions.md
@@ -137,9 +137,19 @@ Participle tags (`v.prap.`, `v.prpp.`, `v.pfap.`, `v.pfpp.`) map to `VerbForm=Pa
 
 ## MISC
 
+The column is typed key by key in `@interslavic/conllu` (`Misc`): an
+attribute not listed there is a type error, not a new string in the column.
+
 | Key | Meaning | Set by |
 |---|---|---|
-| `SID` | Steenbergen id from source row | `inflect()`, passed from caller |
+| `SID` | Identity of the entry: the Steenbergen id of the source row, or `<id>-<suffix>` for a lexeme `derive()` produced | `inflect()`, passed from caller; `derive()` |
+| `LID` | Steenbergen id of the row a derived lexeme descends from | `derive()`, when given a `lid` |
+| `Derivation` | Which rule produced a derived lexeme, as a `DerivationCode` (`v_prap`, `adj_adv`) | `derive()` |
+| `Declension` | `Adj` on a noun that is an adjective used as one (`učeny`, `prěměnna`). UPOS says `NOUN`, so nothing else records which tables the forms came from | `inflect()` |
+| `SpaceAfter` | `No` where the original text had no space | `splitTokens()` |
+
+The golden dump carries sid, lemma, form and FEATS, not MISC, so a change to
+an attribute above moves snapshots and nothing else.
 
 ## Normalization
 

diff --git a/packages/conllu/src/format.test.ts b/packages/conllu/src/format.test.ts
@@ -129,15 +129,20 @@ describe('formatTokenAsConllu', () => {
       );
     });
 
+    test('drops an attribute that arrived as an explicit undefined', () => {
+      const token: Token = { form: 'a', misc: { SID: '1', LID: undefined } };
+      assert.equal(columns(formatTokenAsConllu(token))[9], 'SID=1');
+    });
+
     // `Key=` is not a legal MISC entry, so an empty value drops out entirely.
     test('empty values are dropped', () => {
-      const token: Token = { form: 'a', misc: { SID: '1', Gloss: '' } };
+      const token: Token = { form: 'a', misc: { SID: '1', LID: '' } };
       assert.equal(columns(formatTokenAsConllu(token))[9], 'SID=1');
     });
 
     test('a bag of nothing but empty values is _', () => {
       assert.equal(
-        columns(formatTokenAsConllu({ form: 'a', misc: { Gloss: '' } }))[9],
+        columns(formatTokenAsConllu({ form: 'a', misc: { LID: '' } }))[9],
         '_',
       );
     });

diff --git a/packages/conllu/src/format.ts b/packages/conllu/src/format.ts
@@ -1,9 +1,12 @@
 import type { Token } from './schema.ts';
+import type { Misc } from './misc.ts';
 
 // An empty value would print a bare `Key=`, which is not a legal MISC entry.
-function formatMisc(misc?: Record<string, string>): string {
+// An optional property spread from another bundle can also arrive as an
+// explicit `undefined`; that is an absent attribute, not an empty one.
+function formatMisc(misc?: Misc): string {
   const pairs = Object.entries(misc ?? {})
-    .filter(([, value]) => value !== '')
+    .filter(([, value]) => value != null && value !== '')
     .map(([key, value]) => `${key}=${value}`);
 
   return pairs.length > 0 ? pairs.join('|') : '_';

diff --git a/packages/conllu/src/index.ts b/packages/conllu/src/index.ts
@@ -1,6 +1,7 @@
 export * from './features-atomic.ts';
 export * from './features-pos.ts';
 export * from './format.ts';
+export * from './misc.ts';
 export * from './schema.ts';
 export * from './split-tokens.ts';
 export * from './upos.ts';
diff --git a/packages/conllu/src/misc.ts b/packages/conllu/src/misc.ts
@@ -0,0 +1,60 @@
+/**
+ * The MISC column, spelled out key by key.
+ *
+ * MISC takes any annotation with no home in the standard columns, so left
+ * untyped it accumulates keys nobody declared. Naming every key it may carry
+ * makes adding one a deliberate edit, next to what is already there, and makes
+ * a collision with an existing key a type error.
+ *
+ * @see https://universaldependencies.org/format.html
+ */
+export type Misc = {
+  // Attributes the CoNLL-U format itself defines.
+
+  /** No whitespace between this token and the next one in the original text. */
+  SpaceAfter?: 'No';
+
+  // Attributes of this project.
+
+  /** Steenbergen id of the dictionary entry the token was inflected from. */
+  LID?: string;
+  /**
+   * Identifier of the entry a token belongs to: the `LID` itself for a word
+   * out of the dictionary, and `<LID>-<suffix>` for one `derive` produced.
+   */
+  SID?: string;
+  /**
+   * Which rule produced a derived token. Absent on tokens that are dictionary
+   * entries in their own right, however they inflect.
+   */
+  Derivation?: DerivationCode;
+  /**
+   * `Adj` on a lemma that is an adjective used as a noun. UPOS says `NOUN`,
+   * as the lexicon means it to for `učeny` or `prěměnna`, so this is the only
+   * place the adjectival declension is visible.
+   */
+  Declension?: 'Adj';
+};
+
+/**
+ * What `Derivation` may say: one code per rule that turns a dictionary entry
+ * into a lexeme of its own. The codes live next to the key they fill rather
+ * than beside the rules that use them, so a rule added without a code is a
+ * type error.
+ */
+export type DerivationCode =
+  // Verb derivations
+  | 'v_prap' // Present active participle (dělajųći)
+  | 'v_prpp' // Present passive participle (dělajemy)
+  | 'v_pfap' // Past active participle (sdělavši)
+  | 'v_pfpp' // Past passive participle (dělany)
+  | 'v_noun' // Verbal noun (dělanje)
+  // Adjective derivations
+  | 'adj_cmp' // Comparative adjective, synthetic (dobrějši)
+  | 'adj_sup' // Superlative adjective, synthetic (najdobrějši)
+  | 'adj_sup_a' // Superlative adjective, analytical (najdobry)
+  | 'adj_adv' // Adverb positive (dobro)
+  // Adverb derivations
+  | 'adv_cmp' // Comparative adverb, synthetic (bolje)
+  | 'adv_sup' // Superlative adverb, synthetic (najbolje)
+  | 'adv_sup_a'; // Superlative adverb, analytical (najdobro)
diff --git a/packages/conllu/src/schema.ts b/packages/conllu/src/schema.ts
@@ -1,3 +1,4 @@
+import type { Misc } from './misc.ts';
 import type * as FeaturesPOS from './features-pos.ts';
 import type { UPOS } from './upos.ts';
 
@@ -41,7 +42,7 @@ export interface Token<POS extends UPOS = UPOS> {
   head?: number;
   deprel?: string;
   deps?: Array<{ head: number; deprel: string }>;
-  misc?: Record<string, string>;
+  misc?: Misc;
 }
 
 export type TokenFeatures = FeaturesPOS.Any;
@@ -54,4 +55,6 @@ export type AdapterResult<POS extends UPOS = UPOS> = Array<{
   form: string;
   upos: POS;
   feats: POSFeatures[POS];
+  /** Annotation with no home in FEATS, merged into the token's MISC column. */
+  misc?: Misc;
 }>;
diff --git a/packages/conllu/src/split-tokens.ts b/packages/conllu/src/split-tokens.ts
@@ -1,4 +1,5 @@
 import type { Token } from './schema.ts';
+import type { Misc } from './misc.ts';
 
 type TokenClass = 'word' | 'punct' | 'space' | 'number' | 'dash';
 
@@ -113,7 +114,9 @@ export function splitTokens(str: string): Token[] {
 
   return rawTokens.map((raw, index) => {
     const id = index + 1;
-    const misc = raw.hasSpaceAfter ? undefined : { SpaceAfter: 'No' };
+    const misc: Misc | undefined = raw.hasSpaceAfter
+      ? undefined
+      : { SpaceAfter: 'No' };
 
     if (raw.class === 'punct') {
       return {

diff --git a/packages/morphology/src/derive/derive.ts b/packages/morphology/src/derive/derive.ts
@@ -1,4 +1,4 @@
-import type { Token } from '@interslavic/conllu';
+import type { Misc, Token } from '@interslavic/conllu';
 import type { Lemma } from '../schema.ts';
 import type { KnownForm } from '../inflect/inflect.ts';
 import { normalize, type SteenInput } from '../normalize.ts';
@@ -155,7 +155,7 @@ export function derive(
           feats.Gender = 'Neut';
         }
 
-        const misc: Record<string, string> = {
+        const misc: Misc = {
           Derivation: mapping.code,
         };
 

diff --git a/packages/morphology/src/derive/virtual-id.ts b/packages/morphology/src/derive/virtual-id.ts
@@ -1,24 +1,7 @@
-/**
- * Short canonical derivation identifiers for CoNLL-U misc.Derivation.
- */
-export type DerivationCode =
-  // Verb derivations
-  | 'v_prap' // Present active participle (dělajųći)
-  | 'v_prpp' // Present passive participle (dělajemy)
-  | 'v_pfap' // Past active participle (sdělavši)
-  | 'v_pfpp' // Past passive participle (dělany)
-  | 'v_noun' // Verbal noun (dělanje)
-  // Adjective derivations
-  | 'adj_cmp' // Comparative adjective, synthetic (dobrějši)
-  | 'adj_sup' // Superlative adjective, synthetic (najdobrějši)
-  | 'adj_sup_a' // Superlative adjective, analytical (najdobry)
-  | 'adj_adv' // Adverb positive (dobro)
-  // Adverb derivations
-  | 'adv_cmp' // Comparative adverb, synthetic (bolje)
-  | 'adv_sup' // Superlative adverb, synthetic (najbolje)
-  | 'adv_sup_a' // Superlative adverb, analytical (najdobro)
-  // General word formation
-  | string;
+import type { DerivationCode } from '@interslavic/conllu';
+
+// The codes themselves are declared beside the MISC key they fill.
+export type { DerivationCode };
 
 export type DerivationType =
   | 'prap'

diff --git a/packages/morphology/src/inflect/__tests__/__snapshots__/nouns.test.ts.snap b/packages/morphology/src/inflect/__tests__/__snapshots__/nouns.test.ts.snap
@@ -206,7 +206,7 @@ exports[`conllufy - nouns 20837 1`] = `
 exports[`conllufy - nouns 22830 1`] = `
 "znajema	znajema	NOUN	f.	Case=Nom|Gender=Fem|Number=Sing
 znajeme	znajema	NOUN	f.	Case=Nom|Gender=Fem|Number=Plur
-znajemo	znajema	NOUN	f.	Case=Acc|Gender=Fem|Number=Sing
+znajemų	znajema	NOUN	f.	Case=Acc|Gender=Fem|Number=Sing
 znajeme	znajema	NOUN	f.	Case=Acc|Gender=Fem|Number=Plur
 znajemoj	znajema	NOUN	f.	Case=Gen|Gender=Fem|Number=Sing
 znajemyh	znajema	NOUN	f.	Case=Gen|Gender=Fem|Number=Plur

diff --git a/packages/morphology/src/inflect/__tests__/__snapshots__/numerals.test.ts.snap b/packages/morphology/src/inflect/__tests__/__snapshots__/numerals.test.ts.snap
@@ -46,7 +46,9 @@ exports[`conllufy - numerals 2 1`] = `
 četvŕtině	četvŕtina	NOUN	num.fract.	Case=Dat|Gender=Fem|NumType=Frac|Number=Sing
 četvŕtinam	četvŕtina	NOUN	num.fract.	Case=Dat|Gender=Fem|NumType=Frac|Number=Plur
 četvŕtinojų	četvŕtina	NOUN	num.fract.	Case=Ins|Gender=Fem|NumType=Frac|Number=Sing
-četvŕtinami	četvŕtina	NOUN	num.fract.	Case=Ins|Gender=Fem|NumType=Frac|Number=Plur"
+četvŕtinami	četvŕtina	NOUN	num.fract.	Case=Ins|Gender=Fem|NumType=Frac|Number=Plur
+četvŕtino	četvŕtina	NOUN	num.fract.	Case=Voc|Gender=Fem|NumType=Frac|Number=Sing
+četvŕtiny	četvŕtina	NOUN	num.fract.	Case=Voc|Gender=Fem|NumType=Frac|Number=Plur"
 `;
 
 exports[`conllufy - numerals 3 1`] = `
@@ -95,7 +97,9 @@ exports[`conllufy - numerals 4 1`] = `
 četverkě	četverka	NOUN	num.subst.	Case=Dat|Gender=Fem|Number=Sing
 četverkam	četverka	NOUN	num.subst.	Case=Dat|Gender=Fem|Number=Plur
 četverkojų	četverka	NOUN	num.subst.	Case=Ins|Gender=Fem|Number=Sing
-četverkami	četverka	NOUN	num.subst.	Case=Ins|Gender=Fem|Number=Plur"
+četverkami	četverka	NOUN	num.subst.	Case=Ins|Gender=Fem|Number=Plur
+četverko	četverka	NOUN	num.subst.	Case=Voc|Gender=Fem|Number=Sing
+četverky	četverka	NOUN	num.subst.	Case=Voc|Gender=Fem|Number=Plur"
 `;
 
 exports[`conllufy - numerals 6 1`] = `

diff --git a/packages/morphology/src/inflect/inflect-word.ts b/packages/morphology/src/inflect/inflect-word.ts
@@ -1,5 +1,5 @@
 import { parsePos } from '../partOfSpeech/index.ts';
-import type { AdapterResult, Token, XPOS } from '@interslavic/conllu';
+import type { AdapterResult, Misc, Token, XPOS } from '@interslavic/conllu';
 import { etymologify } from '../etymologify.ts';
 import { parseXPos } from '../steen/index.ts';
 import {
@@ -15,7 +15,7 @@ function composeTokens(
   results: AdapterResult,
   lemma: string,
   xpos: XPOS,
-  misc?: Record<string, string>,
+  misc?: Misc,
 ): Token[] {
   return results.map((result) => ({
     form: etymologify(result.form),
@@ -24,8 +24,9 @@ function composeTokens(
     xpos,
     feats: result.feats,
     // A copy per token: one shared object would let a consumer annotating one
-    // form of the paradigm annotate all the others behind its back.
-    misc: misc && { ...misc },
+    // form of the paradigm annotate all the others behind its back. What the
+    // adapter has to say comes first, so a caller can override it.
+    misc: misc || result.misc ? { ...result.misc, ...misc } : undefined,
   }));
 }
 
@@ -34,7 +35,7 @@ export function inflectWord(
   xpos: XPOS,
   lemma: string,
   irregular?: string,
-  misc?: Record<string, string>,
+  misc?: Misc,
 ): Token[] {
   if (!word || typeof word !== 'string' || word.includes(' ')) {
     return [];

diff --git a/packages/morphology/src/inflect/noun/__tests__/__snapshots__/feminine.test.ts.snap b/packages/morphology/src/inflect/noun/__tests__/__snapshots__/feminine.test.ts.snap
@@ -148264,7 +148264,7 @@ exports[`noun feminine 21351 1`] = `
   "acc": [
     [
       {
-        "form": "narěčeno",
+        "form": "narěčenų",
       },
     ],
     [
@@ -155079,7 +155079,7 @@ exports[`noun feminine 22830 1`] = `
   "acc": [
     [
       {
-        "form": "znajemo",
+        "form": "znajemų",
       },
     ],
     [
@@ -156126,7 +156126,7 @@ exports[`noun feminine 23050 1`] = `
   "acc": [
     [
       {
-        "form": "doråstlo",
+        "form": "doråstlų",
       },
     ],
     [
@@ -156215,7 +156215,7 @@ exports[`noun feminine 23051 1`] = `
   "acc": [
     [
       {
-        "form": "vȯzråstlo",
+        "form": "vȯzråstlų",
       },
     ],
     [
@@ -159526,7 +159526,7 @@ exports[`noun feminine 23429 1`] = `
   "acc": [
     [
       {
-        "form": "zapęto",
+        "form": "zapętų",
       },
     ],
     [
@@ -186832,7 +186832,7 @@ exports[`noun feminine 26879 1`] = `
   "acc": [
     [
       {
-        "form": "krivo",
+        "form": "krivų",
       },
     ],
     [
@@ -235832,7 +235832,7 @@ exports[`noun feminine 35526 1`] = `
   "acc": [
     [
       {
-        "form": "prěměnno",
+        "form": "prěměnnų",
       },
     ],
     [

diff --git a/packages/morphology/src/inflect/noun/__tests__/__snapshots__/neuter.test.ts.snap b/packages/morphology/src/inflect/noun/__tests__/__snapshots__/neuter.test.ts.snap
@@ -40694,7 +40694,7 @@ exports[`noun neuter 12566 1`] = `
   "acc": [
     [
       {
-        "form": "srědnji",
+        "form": "srědnje",
       },
     ],
     [
@@ -72614,7 +72614,7 @@ exports[`noun neuter 23160 1`] = `
   "acc": [
     [
       {
-        "form": "šampansky",
+        "form": "šampansko",
       },
     ],
     [