The paper proposes syntactic n‑grams (sn‑grams), constructed by traversing syntactic tree paths rather than linear word order, and evaluates them as machine‑learning features for natural‑language processing tasks, specifically demonstrating superior performance over traditional word, POS‑tag, and character n‑grams in authorship attribution experiments using SVM, Naïve Bayes, and J48 classifiers.
Bears on: authorship & stylometry · concepts & methods