Анализ зависимостей помогает в нахождении связей между термами используя информацию, предоставленную в [[:Контекстно-свободные_грамматики#Дерево_разбора|деревьях разбора]].
Dependency analysis helps in finding relations between terms by using dependency information present in parsing trees (37). Ciaramita et al. (38) used dependency paths information present in parse trees to find relationship patterns. For two specific concepts, they found relations by extracting the shortest path among those concepts in parsing tree. Their approach was able to learn 83.3% correct relations from corpus. Besides this, it was also used by Sordo et al. (39) as relation extraction technique. Lexico-syntactic pattern is a rule-based approach that plays its role in taxonomic and non-taxonomic relation extraction phases of ontology learning. To extract relations, this algorithm makes use of regular expressions. For example, ‘NP such as NP, NP, … , NP’ is a rule that will extract patterns like ‘seasons, such as summer, winter, autumn, and spring’. This type of rule-based approach is quite helpful in extracting is–a relationship, i.e. is a (summer, season). On the other hand, lexico-syntactic patterns like ‘NP is a part of NP’ can be used to extract non-taxonomic relationships. In 1998, Hearst (40) introduced an algorithm that enabled the extraction of different types of lexico-syntactic patterns. She extracted 106 relations from New York Times corpus in which 61 relations were validated by WordNet. In other words she obtained a minimum accuracy of 75.55%. Besides this, Sombatsrisomboon et al. (149) used these patterns for extraction of taxonomic relations. Buitelaar (15), Kaushik and Chatterjee (41), Ismail et al. (42, 43), Panchenko et al. (44) and Atapattu et al. (45) also used these patterns in their work and concluded that lexico-syntactic patterns provide a reasonably good precision. However, the manual effort required to produce these patterns from data sets is also very extensive. Therefore, Snow et al. (46) made effort in extracting such patterns by using machine learning algorithms. Using logistic regression on a training set of known hypernyms pairs, they automatically learned dependency paths from parse tree and subsequently used them to extract new relationships in unknown data.
Semantic lexicons are knowledge resources in the domain of ontology that play a vital role at different levels of ontology learning (47). Famous semantic lexicons include WordNet (https://wordnet.princeton.edu/) and Unified Medical Language System (https://www.nlm.nih.gov/research/umls/). Semantic lexicons can be used to extract terms, concepts and taxonomic and non-taxonomic relations. They offer a wide range of predefined concepts and relations. These concepts are organized into set of similar words called synsets (sets of synonyms). In (48), Turcato et al. used these synsets for the formation of concepts. Besides this, semantic lexicons also have a number of predefined associations like hypernymy, meronymy etc. They have been employed by Navigli et al. (49) for extraction of taxonomic and non-taxonomic relations.
===Статистические методы===
