Social Media LinguisticsCombining linguistics, NLP, machine learning, and the social sciences to perform social media analysis: detect the main trends and issues discussed online, identify sentiment on those topics, investigate the use of rhetoric and metaphors, and study in-group language features and social group formation.
Tajiki Persian Machine TranslationA finite-state transducer that converts Tajiki Persian text (in Cyrillic) to Iranian Persian script (Perso-Arabic) and runs the resulting transliterated document through an existing Persian-to-English MT system. We use this strategy for the rapid prototyping of MT for the low-resource Tajiki language.
Persian Heritage Language- Developing a language teaching tool for analysis of Persian text, based on morphological analysis.
- Grammar book designed for Persian heritage speakers.
Recent Conference Organization- Third Workshop on Computational Approaches to Arabic Script-based Languages; MT Summit XII, Ottawa, August 2009
- International Conference on Complex Predicates in Iranian Languages; Paris, July 2008
Persian NLP: Shiraz Project (1997-1999)I was the computational linguist responsible for the development of the Shiraz machine translation system at the Computing Research Lab (CRL) in New Mexico State University. The Shiraz project was a MT prototype developed at CRL that translated Persian text into English and used typed feature structures and an underlying unification-based formalism to describe Persian linguistic phenomena. It used an electronic bilingual Persian to English dictionary consisting of approximately 50,000 terms, a complete morphological analyzer, a syntactic parser as well as transfer and generation modules. The system components were tested on a bilingual tagged corpus developed from a large Persian corpus of on-line material (approximately 10MB). The machine translation system is mainly targeted at translating news material.
Coverage: Tokenization and full morphological analysis. Compounds and light verbs were also recognized. The syntactic parser could analyze noun phrases (including relative clauses), preposition phrases and basic sentential constructions. The resulting feature structures were transferred into English syntax and morphological generation was performed on the final translations. The dictionary was built by a team of Persian lexicographers and included single words, compounds and phrasal expressions. It contained information about the orthography, morphosyntactic category and syntactic properties of lexical items as well as the English word-sense equivalents.
Detailed write-ups from the Shiraz project can be found under the publications page: technical reports. However, CRL does not exist anymore and the project components are not available.
Persian NLP: Entity Extraction at Inxight (2002-2004)I was responsible for developing the linguistic aspects of the Persian (Farsi) information extraction system at Inxight Software (now Business Objects). The system performed full segmentation, morphological analysis, part of speech tagging, shallow parsing, named entity extraction, and transcription of proper names to English. The linguistic knowledge is developed with the Xerox finite state technology (XFST) and is disambiguated using an HMM tagger.
For more information, see Inxight/SAP.
Heritage Language Courses TaughtAll classes taught at the University of California San Diego, Linguistics department. Each course consisted of a Culture and Communication session (taught by Elham Sadegholvad) and a language analysis session (taught by me). Full syllabus, exercises and handouts for the courses are available under the Language Analysis site.
Other Computational Projects and Consulting