The task of choosing the correct sense for a word is called word sense disambiguation (WSD). WSD algorithms take an input word w in its context with a fixed set of potential word senses Sw of that input word and produce an output chosen from Sw. In the isolated WSD task, one usually uses the set of senses from a dictionary or theasurus like WordNet.
In the literature, there are actually two variants of the generic WSD task. In the lexical sample task, a small selected set of target words is chosen, along with a set of senses for each target word. For each target word w, a number of corpus sentences (context sentences) are manually labeled with the correct sense of w. In all-words task, systems are given entire sentences and a lexicon with the set of senses for each word in each sentence. Annotators are then asked to disambiguate every word in the text.
In all-words WSD, a classifier is trained to label the words in the text with their set of potential word senses. After giving the sense labels to the words in our training data, the next step is to select a group of features to discriminate different senses for each input word.
The following Table shows an example for the word 'yüz', which can refer to the number '100', to the verb 'swim' or to the noun 'face'.
| Sense | Definition |
|---|---|
| yüz1 (hundred) | The number coming after ninety nine |
| yüz2 (swim) | move or float in water |
| yüz3 (face) | face, visage, countenance |
- Collect a set of sentences to annotate.
- Each sentence in the collection must be named as xxxx.yyyyy in increasing order. For example, the first sentence to be annotated will be 0001.train, the second 0002.train, etc.
- Put the sentences in the same folder such as Turkish-Phrase.
- Build the Java project and put the generated sentence-semantics.jar file into another folder such as Program.
- Put Turkish-Phrase and Program folders into a parent folder.
- Open sentence-semantics.jar file.
- Wait until the data load message is displayed.
- Click Open button in the Project menu.
- Choose a file for annotation from the folder Turkish-Phrase.
- For each word in the sentence, click the word, and choose correct sense for that word.
- Click one of the next buttons to go to other files.
After annotating sentences, you can use DataGenerator package to generate classification dataset for the Word Sense Disambiguation task.
After generating the classification dataset as above, one can use the Classification package to generate machine learning models for the Word Sense Disambiguation task.
You can also see Java, Python, Cython, C++, Swift, or C# repository.
To check if you have a compatible version of Node.js installed, use the following command:
node -v
You can find the latest version of Node.js here.
Install the latest version of Git.
npm install nlptoolkit-wordsensedisambiguation
In order to work on code, create a fork from GitHub page. Use Git for cloning the code to your local or below line for Ubuntu:
git clone <your-fork-git-link>
A directory called util will be created. Or you can use below link for exploring the code:
git clone https://github.com/starlangsoftware/wordsensedisambiguation-js.git
Steps for opening the cloned project:
- Start IDE
- Select File | Open from main menu
- Choose
WordSenseDisambiguation-Jsfile - Select open as project option
- Couple of seconds, dependencies will be downloaded.
In order to sense annotate a parse tree, one can use autoSemantic method of the TurkishSentenceAutoSemantic class.
let sentence = ...
let wordNet = new WordNet();
let fsm = new FsmMorphologicalAnalyzer();
let turkishAutoSemantic = new TurkishSentenceAutoSemantic(wordnet, fsm)
turkishAutoSemantic.autoSemantic()
@INPROCEEDINGS{8093442,
author={O. {Açıkgöz} and A. T. {Gürkan} and B. {Ertopçu} and O. {Topsakal} and B. {Özenç} and A. B. {Kanburoğlu} and İ. {Çam} and B. {Avar} and G. {Ercan} and O. T. {Yıldız}},
booktitle={2017 International Conference on Computer Science and Engineering (UBMK)},
title={All-words word sense disambiguation for Turkish},
year={2017},
volume={},
number={},
pages={490-495},
doi={10.1109/UBMK.2017.8093442}}
- main and types are important when this package will be imported.
"main": "dist/index.js",
"types": "dist/index.d.ts",
- Dependencies should be maximum (not only direct but also indirect references should also be given), everything directly in the code should be given here.
"dependencies": {
"nlptoolkit-corpus": "^1.0.12",
"nlptoolkit-dictionary": "^1.0.14",
"nlptoolkit-morphologicalanalysis": "^1.0.19",
"nlptoolkit-xmlparser": "^1.0.7"
}
- Compiler flags currently includes nodeNext for importing.
"compilerOptions": {
"outDir": "dist",
"module": "nodeNext",
"sourceMap": true,
"noImplicitAny": true,
"removeComments": false,
"declaration": true,
},
- tests, node_modules and dist should be excluded.
"exclude": [
"tests",
"node_modules",
"dist"
]
- Should include all ts classes.
export * from "./CategoryType"
export * from "./InterlingualDependencyType"
export * from "./InterlingualRelation"
export * from "./Literal"
- Add data files to the project folder. Subprojects should include all data files of the parent projects.
- Classes should be defined as exported.
export class JCN extends ICSimilarity{
- Do not forget to comment each function.
/**
* Computes JCN wordnet similarity metric between two synsets.
* @param synSet1 First synset
* @param synSet2 Second synset
* @return JCN wordnet similarity metric between two synsets
*/
computeSimilarity(synSet1: SynSet, synSet2: SynSet): number {
- Function names should follow caml case.
setSynSetId(synSetId: string){
- Write getter and setter methods.
getRelation(index: number): Relation{
setName(name: string){
- Use standard javascript test style.
describe('SimilarityPathTest', function() {
describe('SimilarityPathTest', function() {
it('testComputeSimilarity', function() {
let turkish = new WordNet();
let similarityPath = new SimilarityPath(turkish);
assert.strictEqual(32.0, similarityPath.computeSimilarity(turkish.getSynSetWithId("TUR10-0656390"), turkish.getSynSetWithId("TUR10-0600460")));
assert.strictEqual(13.0, similarityPath.computeSimilarity(turkish.getSynSetWithId("TUR10-0412120"), turkish.getSynSetWithId("TUR10-0755370")));
assert.strictEqual(13.0, similarityPath.computeSimilarity(turkish.getSynSetWithId("TUR10-0195110"), turkish.getSynSetWithId("TUR10-0822980")));
});
});
});
- Enumerated types should be declared with enum.
export enum CategoryType {
MATHEMATICS, SPORT, MUSIC, SLANG, BOTANIC,
PLURAL, MARINE, HISTORY, THEOLOGY, ZOOLOGY,
METAPHOR, PSYCHOLOGY, ASTRONOMY, GEOGRAPHY, GRAMMAR,
MILITARY, PHYSICS, PHILOSOPHY, MEDICAL, THEATER,
ECONOMY, LAW, ANATOMY, GEOMETRY, BUSINESS,
PEDAGOGY, TECHNOLOGY, LOGIC, LITERATURE, CINEMA,
TELEVISION, ARCHITECTURE, TECHNICAL, SOCIOLOGY, BIOLOGY,
CHEMISTRY, GEOLOGY, INFORMATICS, PHYSIOLOGY, METEOROLOGY,
MINERALOGY
}
- If there are multiple constructors for a class, define them as constructor1, constructor2, ..., then from the original constructor call these methods.
constructor1(symbol: any){
constructor2(symbol: any, multipleFile: MultipleFile) {
constructor(symbol: any, multipleFile: MultipleFile = undefined) {
if (multipleFile == undefined){
this.constructor1(symbol);
} else {
this.constructor2(symbol, multipleFile);
}
}
- Importing should be done via import method with referencing the node-modules.
import {Corpus} from "nlptoolkit-corpus/dist/Corpus";
import {Sentence} from "nlptoolkit-corpus/dist/Sentence";
- Use xmlparser package for parsing xml files.
var doc = new XmlDocument("test.xml")
doc.parse()
let root = doc.getFirstChild()
let firstChild = root.getFirstChild()