Converting text and speech in Arabic dialect to structured geographic information—such as GPS coordinates—for domain-specific applications (e. g. , public transport route planning) remains an open research problem in computational linguistics. This report documents the design of a complete NLP pipeline that transforms Syrian dialectal Arabic text or voice queries into structured geographic coordinates that can be fed to an external route planner, as part of a public transit application for Damascus, Syria. The design is motivated by the inadequacy of existing geocoding services, which do not handle dialectal input, local place nicknames, or the typographic variations common in user queries. The pipeline comprises five key components: (1) an Arabic text normalization module based on CAMeL Tools, (2) a named entity recognition (NER) component that identifies ORIGIN and DESTINATION entities with four custom types (AREA, LANDMARK, RELATION, MICRODIRECTION), (3) a hierarchical geocoding engine powered by a curated PostGIS place database, (4) a confidence-based disambiguation interface, and (5) a REST API that exposes the entire pipeline as a Python FastAPI microservice. The system targets an NER F1-score of at least 78%, place disambiguation rank-1 accuracy of at least 70%, rank-3 accuracy of at least 88%, and API response time under 300 milliseconds at the 95th percentile. The proposed system operates entirely locally, without reliance on cloud services, ensuring user privacy and cost-free usage. This work was conducted at Arab International University (AIU), Syria. The official website of the university is: https: //www. aiu. edu. sy
Ghaith Khawly (Sun,) studied this question.