Arabic nested noun compound extraction is the task of identifying the series of two or more attached Arabic nouns existed in certain contexts. The literature has showed a great interests in using statistical measures for performing such task. However, such measures are mainly relying on the co-occurrence of noun compounds within a predefined corpus where the rare ones would have low probabilities for being extracted. Taking the advantage of modern text representation such as the word embedding, this paper aims to propose a combination of convolutional neural network (CNN) and Long Short Term Memory (LSTM) architectures to perform the extraction of Arabic noun compounds. CNN is intended to generate embedding for the words, while the LSTM is intended to conduct the classification where the words will be categorized into noun compounds and non-noun compounds. A benchmark Arabic corpus has been used in the experiments. Results showed that the proposed combination has extracted the nested noun compounds with an average precision of 0.905. This result indicates the usefulness of using word embedding for the Arabic nested noun compounds extraction.
Volume 10 | Issue 11
Pages: 488-495