Skip to content

Releases: espositopasquale/Nnois

Release list

Zero release

Choose a tag to compare

@espositopasquale espositopasquale released this 01 Sep 09:59

Nnois 0.1.0

First Public Release

Nnois is a Java command-line tool for multilingual orthographic transparency analysis. This first release provides context-sensitive lexical comparison: it identifies a likely sense for a source word or nominal unit, retrieves forms associated with the same concept in other languages, and ranks their normalized spelling similarity.

Included

  • Chosen-word analysis: compare a user-selected word in sentence context with lexicalizations of its selected sense in other languages.
  • Aggregate nominal analysis: extract nouns and validated noun phrases from a text, resolve their senses, and sum language-level transparency contributions.
  • Aggregate detail report: inspect the unit-level lemma and score behind an aggregate result.
  • Single-word sense explorer: browse available senses, glosses, and multilingual translations for an isolated word.
  • Language selection: use automatic language detection or enter a supported ISO 639-1 or ISO 639-2 code manually.
  • Terminal-oriented UX: recover from invalid menu choices, return to text entry with back, and display language labels with ANSI colors inspired by associated flags.
  • Linguistic processing: tokenization, POS tagging, lemmatization, noun-phrase chunking, and Lesk-style contextual sense disambiguation when language resources are available.

Method At A Glance

Nnois calculates normalized Levenshtein similarity between a source form $s$ and target lemma $t$:

$$ T(s,t) = 1 - \frac{d(s,t)}{\max(|s|, |t|)} $$

The source and target forms are lowercased and trimmed before comparison. Sense selection is constrained by context using overlap between usable context terms and candidate gloss terms. Aggregate results add the contributions from individual nominal units, so they are sums rather than probabilities and can exceed 1.00.

Important Interpretation Note

Nnois measures orthographic similarity under a shared sense constraint. It can surface visibly similar, cognate-like forms for exploration, but it does not prove historical cognacy and does not measure pronunciation, translation quality, word frequency, or language-learning difficulty.

Requirements And Run

  • Java 21 or a compatible JDK
  • Maven 3.8 or later
cd nnois
mvn clean package
mvn exec:java -Dexec.mainClass=com.nnois.App

License And Attribution

The Nnois source code is licensed under the GNU General Public License. Linguistic data, language models, and software dependencies included or used by the project remain subject to their own license terms and notices, summarized in Attributions.md.

Known Scope

  • Analysis quality depends on the coverage and quality of installed linguistic models and lexical data.
  • The application uses fallbacks where language-specific NLP models are unavailable.
  • The detailed report explains aggregate contributions; it does not turn raw aggregate scores into length-normalized averages.

For full usage instructions, methodology, supported language-code mappings, and troubleshooting, see the README.