Like other machine learning applications, one major problem facing ontology learning is how to estimate the probability of co-occurrences that are not observed in a training corpus. Such a data-sparseness problem can be addressed with two types of approaches: smoothing and class-based methods. Smoothing methods estimate the probability of unobserved co-occurrences using observed frequency information [61, 62]. Class based models [63] distinguish between unobserved co-occurrences using classes of ‘‘similar’’ words. The probability of a specific co-occurrence is determined using generalized parameters about the probability of class co-occurrences. Ontology learning techniques have been applied to a variety of domains, which could have impact on the selection of ontology learning techniques.