The Arabic language today is a collection of varieties that are historically related to Classical Arabic. The standard variety, Modern Standard Arabic (MSA), is the accepted primarily written language of culture, media and education throughout the Arab World. The primarily spoken non-standardized varieties, or dialects, not only vary among themselves, but are also different from MSA. The differences are lexical, phonological, morphological and to a lesser degree syntactic. These differences make the direct use of MSA natural language processing (NLP) tools and applications for handling dialects impractical. Linguistic studies of Arabic dialects are often limited to small-scale data collection. In the field of NLP, a few Arabic dialects have started receiving more attention, particularly in speech recognition and machine translation. While our ultimate goal is to develop robust NLP technologies to handle all forms of Arabic, standard and dialectal, this proposal focuses on Arabic dialects. The proposal has two parts: resources and applications. For resources, we plan to build a suite of four large-scale fine-grained multi-dialectal resources. First there will be a lexicon covering the most commonly used concepts with their expressed vocabulary in 25 Arab cities spread across the Arab World. Second, we plan to collect a corpus of over 110,000 sentences representing the different Arabic dialects spoken across the 25 cities; the corpus will be created by translating parts of a multilingual corpus of travel expressions into different dialects. Third, the entries in the lexicons and some of the corpus sentences will be linked to an atlas indicating where they are used or accepted; the atlas will be created through a large crowdsourcing effort targeting hundreds of people in the 25 Arab cities. Finally, we will create a multi-dialectal morphological analyzer. For applications, we propose to conduct research on two important basic technologies. The first is dialect identification at a city-dialect granularity level. The second is robust machine translation tools for translating from the dialects to and from MSA and English. The unique resources we plan to create are necessary for conducting the research we plan to do on applications. Both resources and applications, in addition to developed insights and methods have important implications for research on Arabic NLP specifically and multilingual multidialectal research for other languages.