Development build. This page describes main, not a released package. The latest published Lodestar.Text is 0.6.0 — read its documentation.

Home › Text › Stemming

ArabicSnowballStemmer

Arabic stemming by the Snowball algorithm.

public static class ArabicSnowballStemmer

Example — a noun, the same noun with the definite article, and with a possessive, on one key.

using Lodestar.Text.Stemming;

string bare = ArabicSnowballStemmer.Stem("كتاب");  // => كتاب
string defined = ArabicSnowballStemmer.Stem("الكتاب");  // => كتاب
string possessive = ArabicSnowballStemmer.Stem("كتابها");  // => كتاب

Remarks — Arabic is the odd one of the sixteen, and deliberately so: #176 took it last because its algorithm shares least with the others. Three things follow from that.

It uses neither R1 nor R2. Every other stemmer here computes a region from the first vowel-then-consonant boundary and asks whether a suffix lies inside it. The published Arabic algorithm has no region at all — each rule carries an explicit character count instead, and a four-letter word is simply refused a rule a five-letter word gets. This is why ArabicSnowballStemmer is the one stemmer that does not derive from the shared worker base; see decision 0006.

Normalisation runs first, and it is half the algorithm. The vocalisation marks and the kasheeda are dropped, the lam-alef ligatures are written back as two letters, and the Arabic-Indic digits become the ASCII ones — so مُحَمَّد and محمد are the same word to the rules, and ١٢٣ comes back as 123. The hamza carriers are resolved last instead: they survive the affix steps and then fold, an alef-hamza becoming a bare alef inside a word and a lone hamza at the end of one.

Affixes come off both ends. The suffixes go first — possessives, then case and verb endings, then the nisba ي — and the prefixes after: the conjunctions و and ف when they do not sit in front of an alef, the definite article alone or behind a preposition, and the future markers, whose س is dropped. A word carrying the article is read as a defined noun, and its possessive endings are left alone because the article is what identifies it.

Reference behaviour is snowballstemmer.stemmer("arabic") rather than nltk — the only other language whose corpus is frozen that way is Hungarian, and decision 0006 has the measurement. Matched over 146 words.

Applies to — net10.0, netstandard2.0.

See also — the stemming index.

Members

Member	What it does
ArabicSnowballStemmer.Stem	The Snowball stem of one Arabic word.