Linguistics

In Grammario→

Turkish words are split into a root and labeled suffixes, so a learner sees how bilmediysen is assembled instead of memorizing it whole.

In Lociros→

Sudachi and pymorphy3 reduce every word to its lemma and inflection. That analysis is what the level validator checks and what the reader shows when you tap a word.

In Grammario→

spaCy, with Stanza as a fallback, parses each sentence into a dependency tree before any model is involved. Click a word for its role, lemma, and features.

In Lociros→

Constructions a level forbids, like Japanese relative clauses or Russian который-clauses, are found in the analysis rather than guessed from the text.

In Grammario→

Agreement clusters group the words that must match in Italian and Spanish, governance shows which word demands which case in German and Russian, and rule-based checks flag a mismatch.

In Lociros→

Cases, participles, and moods are listed per level, so an A2 Russian passage only uses the cases an A2 reader has met.

In Grammario→

Each language family gets its own analysis strategy instead of one engine for all six languages.

In Lociros→

Each language gets its own analyzer: Sudachi for Japanese, CAMeL Tools for Arabic, spaCy for Italian, and razdel with pymorphy3 for Russian.

In Lociros→

Tapping a Japanese word aligns each reading to its kanji, with stroke order from KanjiVG. Tapping an Arabic word shows its root and verb form.

In Grammario→

Japanese analysis covers verb and adjective conjugation and honorific register, which a word-by-word gloss would miss.

In Lociros→

Each level’s allowed grammar lives in JSON per language. A validator checks every generated passage and sends it back for a rewrite when it goes over level.

In Grammario→

Sentences get a CEFR difficulty score from features of the parse, like tree depth, subordination, and morphological complexity.