[Paper Review] Persian Wordnet Construction using Supervised Learning

Zahra Mousavi, Heshaam Faili|arXiv (Cornell University)|Apr 11, 2017

Natural Language Processing Techniques3 citations

TL;DR

This paper presents a supervised machine learning approach to automatically construct a Persian Wordnet by linking Persian words to Princeton WordNet synsets using a bilingual dictionary and a Persian corpus. Trained on FarsNet as positive instances, a classifier with seven features achieves 91.18% precision, resulting in a wordnet with over 16,000 words and 22,000 synsets—representing state-of-the-art performance for Persian wordnet construction.

ABSTRACT

This paper presents an automated supervised method for Persian wordnet construction. Using a Persian corpus and a bi-lingual dictionary, the initial links between Persian words and Princeton WordNet synsets have been generated. These links will be discriminated later as correct or incorrect by employing seven features in a trained classification system. The whole method is just a classification system, which has been trained on a train set containing FarsNet as a set of correct instances. State of the art results on the automatically derived Persian wordnet is achieved. The resulted wordnet with a precision of 91.18% includes more than 16,000 words and 22,000 synsets.

Motivation & Objective

• To address the lack of large-scale, high-precision lexical resources for Persian by constructing a comprehensive Persian Wordnet.
• To automate the alignment of Persian words with synsets from Princeton WordNet using supervised learning.
• To improve the efficiency and scalability of Persian Wordnet construction beyond manual or rule-based methods.
• To achieve state-of-the-art performance in automatic Persian Wordnet construction using a trainable classification system.

Proposed method

• Initial word-synset links are generated using a Persian corpus and a bilingual dictionary to map Persian words to Princeton WordNet synsets.
• Seven linguistic and distributional features are extracted to represent each candidate word-synset pair.
• A supervised classification model is trained on a labeled training set where FarsNet serves as the gold standard for correct links.
• The trained classifier distinguishes correct from incorrect word-synset mappings, refining the initial set of links.
• The method relies on feature engineering and supervised learning to minimize human annotation effort.
• The final wordnet is constructed by filtering the initial links using the trained classifier’s predictions.

Experimental results

Research questions

• RQ1 Can a supervised learning approach effectively automate the construction of a Persian Wordnet with high precision?
• RQ2 How well can a classifier trained on FarsNet generalize to new, unseen Persian word-synset pairs?
• RQ3 What set of features yields optimal performance in distinguishing correct from incorrect word-synset alignments in Persian?
• RQ4 To what extent can a bilingual dictionary and a monolingual corpus support automatic wordnet construction without extensive manual labeling?
• RQ5 What is the achievable precision and coverage of an automatically constructed Persian Wordnet using this method?

Key findings

• The proposed method achieves a precision of 91.18% on the automatically constructed Persian Wordnet.
• The final wordnet contains more than 16,000 words and 22,000 synsets, demonstrating high coverage.
• The use of seven handcrafted features in a supervised classifier significantly improves alignment accuracy over baseline methods.
• The system outperforms previous approaches, establishing a new state of the art for Persian Wordnet construction.
• FarsNet serves as an effective gold standard for training the classifier, enabling high-quality automatic expansion.
• The method significantly reduces the need for manual annotation, enabling scalable and efficient wordnet construction.