dumling

dumling is a UD-inspired, learner-facing, meaning-focused lexical classification framework.

It provides a language-independent structure for answering questions like this:

Teacher, what is [give] in:

• “See,” he explained to Aunt Petunia through a mouthful of nails, “if they can’t deliver them they’ll just [give] up.”?

For a learner, the useful answer is not only that [give] is a verb token.

dumling-way is to classifies [give] as:

• a Partial Selection of the normalized full
• Surface “give up”, which points to the
• Lemma “give up”: a phrasal English VERB

That is the core job of dumling: keep the learner’s highlighted text connected to the full lexical item that carries the meaning.

Why It Exists

Classical linguistic segmenters are usually built around tokens and grammar. That is useful, but it is too granular for many learner-facing explanations.

In a UD-style analysis, “give up” is split into parts: give is the verbal head, and up is a particle attached to it. Dumling can still borrow that POS vocabulary, but it also gives applications a way to treat “give up” as one lexical unit when that is the meaning-bearing unit the learner needs.

This also matters for phrasemes:

• During my [walk] in a park, I saw a squirrel.

  Here dumdict classifies it as a Full Selection of the noun Lexeme “walk”.

• This exam was a [walk] in the park.

  Here dumdict classifies it as a Partial Selection of the idiomatic Phraseme “walk in the park”.

The spelling on the page is not forced to be the whole lexical object. A small learner highlight can still point to the larger surface and lemma that explain the meaning.

What It Focuses On

Dumling focuses on lexical classification for learning tools:

• Lemma: the lexical item behind the expression, such as “walk”, “give up”, or “walk in the park”
• Surface: the normalized full form in context, such as “gave up”
• Selection: the exact text the learner highlighted, such as give, up, or walk
• lexical kind: Lexeme, Morpheme, Phraseme, or Construction
• learner-relevant selection features: partial coverage, licensed spelling variants, and typos when those facts are marked
• learner-relevant surface features: properties of the resolved full surface, such as archaic status
• language-specific lexical inventories built on a shared cross-language model

Construction is the learner-facing branch for patterned entries such as fused forms like German zum and paired frames such as um zu or entweder oder.

The framework is implemented as a TypeScript and Zod package so apps can validate, serialize, search, and round-trip these objects through IDs.

What It Omits

Dumling intentionally does not try to be a full grammar model.

It does not model syntactic dependency relations, phrase structure, or sentence-level grammar. It does not replace UD treebanks, parsers, or tokenizers. It also does not try to explain every grammatical relation between words in a sentence.

The scope is narrower: identify the meaning-bearing lexical unit behind a learner’s selection, and describe that unit in a stable, language-aware shape.

Start Here

• Linguistics explains the domain terms without TypeScript.
• Model describes the DTO shape: Lemma, Surface, and Selection.
• API lists the package entrypoints and language-bound helpers.
• English, German, and Hebrew summarize the implemented language packs.

Every docs route is built as static HTML and also emitted as a sibling Markdown file. For example, this page is available as /index.html and /index.md.

Runtime Scope

The current implemented runtime languages are:

• en: English
• de: German
• he: Hebrew

The language inventory is curated by the package. Consumers can choose a supported language dynamically with getLanguageApi(language), but arbitrary user-defined language packs are not part of the public runtime API.