Research on WordNet 4: the application field of WordNet

Research on WordNet 4: the application field of WordNet

1. As a synonym for WordNet ·

WordNet is similar to the synonym forest: it also uses the synonym set as the basic construction unit for organization. If you have a known concept in your mind, you can find a suitable word in the synonym set to express this concept. However, WordNet is not just a collection of synonyms. A synonym set is associated with a certain number of relational types. These relationships include the upper-right relationship, the overall part relationship, and the inheritance relationship.

2. WordNet as a general dictionary

Similar to traditional dictionaries, WordNet defines a synonym set and provides examples. Include the definition of these synonyms in the synonym set. The appropriate example sentences are given for different words in a synonym set to distinguish them.

3. Describe the relationship between words

The types of semantic relations in different syntactic parts are also different. For example, although nouns and vertices both form hierarchical semantic relations between words, in terms of nouns, the upper-lower relationship is a hyponymy relationship, the verb contains the troponymy relation, and the entailment (inheritance) relation in the verb is similar to the meronymy (overall) relation in the noun. The meronymy relationship of nouns is divided into three types of subrelationships (see "Nouns in WordNet ).

4. Get the word and its context

to provide contextual information of words, Princeton (Princeton) the Cognitive Science Laboratory has developed a semantic search tool (semantic concordance ). Chapter 2 of the book T. This tool makes up a whole database of text and dictionary, so that words in the text are associated with the appropriate meaning in the dictionary. Such a semantic search tool can be viewed as a text, where words are labeled with syntaxes and semantic information, or as a dictionary, all entries have examples that indicate the definition usage environment. The text that works with the WordNet Semantic Dictionary comes from the corpus of the Brown Corpus (Standard corpus of contemporary American English) and the full text of a short story (the complete text of a novella ).

5. Meaning discrimination

Although it is clear that in the context of determination, the process of discrimination is not easy, the author gives the definition of a synonym. For computers, the context of discrimination is a big problem.

· Leaco CK and chodorow (See Chapter 11th in WordNet) test different strategies for determining the ambiguity of the synonym "Serve. In the three experiments, they found that the "window" of the context is suitable for 6 words, and the result is the best. In addition, when the context information is used together with the semantic similarity information of words in WordNet, the highest accuracy of discrimination is achieved.

 

6. Information Retrieval

Meaning discrimination is a key factor for many applications, such as information retrieval. Voorhees (See Chapter 12th in WordNet) explains that you need to find the required documents in a large number of documents, the computer must effectively match the words in the query with the document title or abstract. Voorhees explores the effectiveness of WordNet in terms of word matching, and finds that difficulties in distinguishing meaning hinders the effective use of the semantic information in WordNet. The Semantic Link information in WordNet is helpful for improving the search results only when the concepts are manually selected to make the words to be searched have a known meaning.

7. semantic relationship and text consistency

Hirst and St-onge (See Chapter 13th of WordNet) also discuss context, especially how a coherent text is made up. Based on the assumption that discourse is linked up by meaning-related concepts, they use the lexical chain concept as a way to evaluate coherence. Hirst and St-onge use Word chains to check malapropism ). They incorrectly define words as: the concepts corresponding to a word are irrelevant to those corresponding to other words in the text where the word is located. Using the method of evaluating the link strength in a Word chain, Hirst and St-onge believe that the greater the semantic distance between words in a text, the more likely a word error may occur.

· Al-Halimi and kazman are also interested in information storage, indexing, and retrieval (See Chapter 14th of WordNet ). They described a method for automatically indexing video conferencing scripts by topic (not by keyword index), and used the topic index results to retrieve information through matching scripts. Al-Halimi and kazman describe topic information as the lexical tree, which is a correction to the Word chain. One of the innovations of the former for the latter is to consider the information relevance of different semantic relationship types. · Hirst and St-onge indicate that WordNet lacks information about the semantic distance between two related words. Their example is: More stew than steak (meat than steak), where "more... than" is a format used to connect two semantic words. In this example, two nouns (stew and steak) belong to six synonym sets (synset), which obviously cannot reflect their actual semantic distance. English speakers know that the two upper and lower spaces of "good person" (good person, SAGE) are semantically similar. The two upper and lower bitwise concepts are {saint, holy man, holy person, Angel}, {plaster saint }, furthermore, the similarity between these two concepts is different from that between them and the third sub-concepts. The third concept is {square shooter, straight arrow ).

 

8. Knowledge Engineering

in many WordNet applications, perhaps the most ambitious is knowledge engineering (See Chapter 15 and 16 in WordNet ).

· harabagiu and Moldovan (see chapter 16th in WordNet) point out that modeling for common sense reasoning requires an extended knowledge base, including a large number of concepts and relationships. WordNet provides the former, but it does not support reasoning in terms of relations. Their solution is to differentiate comments in WordNet to obtain more relationships between words, so as to convert comments in WordNet into semantic networks, which include relations between different word classes. They gave an example: there is a path between hungry (Ele. Me) and refrigerator (refrigerator), because the two markup words collided on the Food node, that is, through food, hungry and refrigerator can be associated for common sense reasoning.