Arabic-English Google Translation Evaluation and Arabic Sentiment Analysis

Machine translation is one of the most important tasks in the automatic processing of the natural languages, but its systems are still very far from achieving any performance close to ideal human translation due to many obstacles and difficulties. For example, grammatical rules between different languages are different, also, some words in the source language may not have an equivalent translation in the target language and this will lead to accurate translation results. Also, sentiment analysis is another application of Natural Language Processing (NLP) which is the process of identifying the feelings or sentiments held in a piece of text and classifying them into positive, negative, or neutral. This study sets out to examine the efficiency and accuracy of Google Translate’s translation between the Arabic and English languages and to evaluate the qualification of sentiment analysis when we apply it to Arabic texts. My study will also examine how old texts and texts written in dialects of Arabic affect both sentiment analysis and Google Translate’s performance. So, I used Arabic songs’ lyrics which differ between each other in terms of precedence and formality where they can be old or new, and formal or informal. To evaluate the results of Google Translate in this thesis, a lexical and grammatical analysis was used and the retrieval rate of words, sentences, and meanings of the text in the source language was calculated. In this thesis, I proposed a new approach to sentiment analysis which I conducted over five Arabic songs’ lyrics. Some of the lyrics are written in informal colloquial Arabic and some in Modern Standard Arabic (MSA). I performed sentiment analysis on those lyrics to evaluate the efficiency of the Google translation by comparing the results of the sentiment analysis of Arabic lyrics and the translated versions of those lyrics. The results of the evaluation of Google Translate showed that Google Translate has failed to achieve adequate translation, especially in old complex informal and unstructured Arabic texts. In addition, the results of applying sentiment analysis to Arabic texts showed that old and informal texts lower the performance of sentiment analysis. Also, applying English sentiment analysis to the translated version gave different results than applying Arabic opinion mining to the Arabic original version. Furthermore, Google Translate lowered the accuracy of the English sentiment analysis compared to the results I got when I applied the English sentiment analysis to my suggested translation of the lyrics.