/felig-toolkit

A toolset for Amharic Language pre-processing. Includes an Amharic Stemmer, Transliterator, Stopword remover , Lexical analyzer, Corpus indexer and Term weighter.

Primary LanguageTypeScriptMIT LicenseMIT

Felig logo

Felig Toolkit

A toolset for Amharic Language pre-processing πŸ”§

Felig Toolkit Web​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​

Now with Typescript support!


What is felig-toolkit?

It is a toolset for Amharic Language pre-processing. It includes an Amharic Stemmer, Amharic Transliterator, Amharic Stopword remover, Amharic Lexical analyzer, Amharic Corpus indexer and Term weighter.

Amharic Lexical Analyzer

Breaks down Amharic language corpus and returns tokens by removing any whitespace, expanding abbreviations(አ.አ -> αŠ α‹²αˆ΅ αŠ α‰ α‰ ), removing numbers, breaking up hyphenated words, and removing punctuation (ፑ ፒ ! ? ...).

Amharic Stopword remover

Removes commonly occuring words that have no contribution to the semantics of the corpus. Eg: αŠ₯αŠ“ ፑ αˆ΅αˆˆα‹šαˆ… ፑ α‰ αˆ˜αˆ†αŠ‘αˆ...

Amharic Transliterator

Changes Unicode Amharic characters to ASCII. Exmaple: αˆαŒ†α‰½ -> αˆαŒ…αŠ¦α‰½ -> ljoc. This tool implements two types of Amharic transliteration lookup tables.

  • SERA (System for Ethiopic Representation in ASCII) - This system maps alphabets with similar sounds separately. Eg: (αˆ€α£αˆα‘αŠ€)፣(ሰፑሠ)ፑ(αŒΈα‘α€)ፑ(α‹α‘αŠ ). However, in practice, these alphabets are used interchangeably and use of SERA would greatly decrease recall. NOT RECOMMENDED!

  • Felig - Normalizes the redundant symbols into a common symbol. RECOMMENDED!

Amharic Stemmer LIVE DEMO

Reduces the different morphological (e.g. inflectional or derivational) variations of Amharic word forms by taking an Amharic word and returning the stem through affix-removal with longest match.

Exmaple: αˆαŒ†α‰½ -> αˆαŒ…αŠ¦α‰½ -> ljoc -> lj -> αˆαŒ…

Amharic Corpus Indexer

Produces an index file for the stemmed words in a corpus and relates them with the files they are found in. It also stores their frequencies per file.

Term Weighter

Calculates the weight of words from the index file using product of their length normalized Term frequency and Inverse document frequency (tf*idf).

Installation

Felig Toolkit is available as a package on NPM for use in a Node application:

# NPM
npm install felig-toolkit
# YARN
yarn add felig-toolkit
# PNPM
pnpm install felig-toolkit

Example

note: this package uses es-modules

import felig_toolkit from 'felig-toolkit'

What's Included

  • felig_transliterate(word,lang): takes a single word and its' language (am/en) and returns felig-transliterated string

  • sera_transliterate(word,lang): takes a single word and its' language (am/en) and returns SERA-transliterated string.

  • rmvStopwrd(corpus): takes an Amharic corpus text (sentence/paragraph/multiple-paragraphs) and removes stop wprds

  • lexAnalyze(corpus): takes an Amharic corpus text returns a string of tokens

  • stem(word): takes an Amharic word string and returns the stem as a string (async)

  • indexer(filesArray, outputIndexFilePath, type): takes an array of files and produces an index (.json) file. (type= "doc" | "query")

  • weigh_terms(indexFilePath, outputWeightedTermsPath, typeOfIndex): takes an index file and produces a file (.json) with weighted terms. (typeOfIndex= "doc" | "query")

How to use in Web apps

Felig toolkit does not work in the browser (requires node.js enviroment).

Use felig-toolkit on your server.

Exmaple: If you are using Next.js, you can use felig-toolkit in a Next server route handler (/api/felig/route.ts) and pass the results.

Contributions

felig-toolkit is open to contributions, but it is recommend to create an issue or reply in a comment to let others know what you are working on first.

How to run locally

Prerequisites

  1. Clone the repository
  2. Run npm install
  3. Run node index.js on the root directory

Attribution

To prepare the following tools, these academic papers were used