Introduction to Word Sense Disambiguation (WSD) (Part One)
Did you know that the versatility of language can lead to multiple interpretations of the same word? Take the word “bass,” for example. In one sentence it can mean a deep musical sound, while in another it refers to a type of fish. Our minds easily switch between these meanings, but for machines, processing unstructured text is not so simple. They must analyze and transform the information into structured data to understand its meaning. The process of machine being able to analyze and transform data to understand its meaning is called Word Sense Disambiguation(WSD).
To further clarify, Word Sense Disambiguation refers to automatically recognizing multiple meaning of ambiguous words recognized in a sentence. The main goal is to tell us the meaning of word that is been used. WSD typically involves two main tasks. determining the different possible senses (or meanings) of each word, tagging each word of a text with its appropriate sense with high accuracy and efficiency.
Applications of Word Sense Disambiguation:
Speaking of the application of Word, it can be applied to the following
- Machine Translation
2. WSD is necessary for retrieving information
3. It is useful for semantic parsing:
4. It can also be used for speech synthesis and recognition
Methods of Word Sense Disambiguation:
There are three main methods of word sense disambiguation. These are:
Knowledge-based Methods: Knowledge-based approaches based on different knowledge sources as machine readable dictionaries or sense inventories, thesauri etc. The common Knowledge based methods are:
a. Lesk Algorithm
b. Semantic Similarity
c. Selectional Preferences
d. Heuristic Method: This method is in three parts: Most Frequent Sense, One Sense per Discourse, One Sense per Collocation
Supervised Method: The supervised approaches applied to WSD systems use machine-learning technique from manually created sense-annotated data. Training set will be used for classifier to learn and this training set consist examples related to target word. The main Supervised Methods are:
a. Decision List
b. Decision Tree
c. Naïve Bayes
d. Neural Networks
e. Exemplar-Based or Instance-Based Learning
a. Support Vector Machine
b. Ensemble Methods: The common ensemble methods are Majority Voting, Probability Mixture, Rank-Based Combination, AdaBoost
Unsupervised Method: These methods do not depend on external knowledge sources or sense inventories, machine readable dictionaries or sense-annotated data set. These algorithms generally do not assign meaning to the words instead they discriminate the word meanings based on information, found in un-annotated corpora. Main approaches of unsupervised are as follow:
a. Context Clustering
b. Word Clustering
c. Co-occurrence Graph
d. Spanning tree based approach
This article is a short that introduced us to the meaning of Word Sense Disambiguation. The next article will explain the implementation of Lesk algorithm to disambiguate words in python programming language.
Watch Out for Part Two Soon
Further Read
