Linguistics

dumling models learner-facing linguistic annotation. The package does not try to be a full grammar engine. It gives applications stable objects for the parts of annotation that commonly need to be validated, serialized, searched, and shown back to learners.

Lemma

A Lemma is the dictionary entry, or lemma-like entry, behind an observed form.

For a word like Seen, the German lemma might be See. For an English form like ran, the lemma might be run.

The lemma owns properties that belong to the lemma entry itself:

language: the concrete language, such as de, en, or he
canonicalLemma: the canonical lemma spelling
lemmaKind: the broad class: Lexeme, Morpheme, Phraseme, or Construction
lemmaSubKind: the concrete subtype, such as NOUN, VERB, Prefix, or Idiom
inherentFeatures: features that belong to the lemma as a lexical item
meaningInEmojis: a compact meaning hint for learner-facing UI

Construction is the public branch for learner-relevant patterned entries such as fused forms like German zum, zur, beim, or ins, and paired frames such as um zu or entweder oder.

Surface

A Surface is the normalized full form in context.

The surface always contains a Lemma. This keeps a normalized word form connected to the lemma entry it realizes.

There are two surface kinds:

Citation: the surface is the citation form itself
Inflection: the surface is an inflected form of the lemma

Inflection surfaces carry inflectionalFeatures, such as number, case, tense, person, gender, degree, definiteness, or verb form, depending on the language and lemma subtype.

Selection

A Selection is the exact string the learner highlighted.

The selection always contains a Surface, and that surface always contains a Lemma. This gives the full chain:

Selection -> Surface -> Lemma

Marked selection mismatches live in the optional selectionFeatures bag:

Field	Marked value	Meaning
`orthography`	`Typo`	the observed selection is misspelled
`coverage`	`Partial`	the learner highlighted only part of the resolved surface
`spelling`	`Variant`	the observed spelling is licensed but non-canonical

Absence means the unmarked default: standard orthography, full coverage, and canonical spelling.

These axes remain independent. A typo can still be partial, and a partial selection can still point to a full normalized surface.

Lemma Kinds

lemmaKind has four values:

Kind	Use
`Lexeme`	words and word-like lexical entries, categorized with Universal Dependencies-style POS tags
`Morpheme`	roots, prefixes, suffixes, clitics, and related sub-word units
`Phraseme`	multi-word or formulaic expressions such as idioms and proverbs
`Construction`	learner-relevant patterned entries such as fused forms like `zum` and paired frames like `um zu`

lemmaSubKind is the public subtype field for all four families. The package does not expose separate public discriminator names like pos, morphemeKind, or phrasemeKind.

Features

Features are split by where they belong:

inherentFeatures describe the lemma itself
inflectionalFeatures describe a concrete inflected surface
selectionFeatures describe how the attested highlight differs from the resolved surface
surfaceFeatures describe marked properties of the resolved surface itself, such as historicalStatus: "Archaic"

Each language narrows the abstract feature inventory. For example, German nouns support grammatical gender as an inherent feature and case/number as inflectional features. English nouns support number inflection but not grammatical case in the same way. Hebrew supports language-specific features such as hebBinyan for verbs.