Encyclopedia of Computer Science and Technology

278 linguistics and computingIn his research activities, Licklider focused his effortsnot so much on AI as on the development of interactivecomputer systems that could promote his vision of humancomputersymbiosis. This included time-sharing systems,where many users could share a large computer system,and networks that would allow users on different computersto communicate with one another. He believed that thecooperative efforts of researchers and programmers coulddevelop complex programs more quickly than teams limitedto a single agency or corporation (see also open-sourcemovement).Licklider’s efforts to focus ARPA’s resources on networkingand human-computer interaction would providethe resources and training that would, in the late 1960s,begin the development of what would become the Internet.Licklider spent the last two decades of his career teachingat MIT. Before his death in 1990, he presciently predictedthat by 2000 people around the world would be linked in aglobal computer network.Further Reading“Internet Pioneers: J. C. R. Licklider.” Available online. URL: http://www.ibiblio.org/pioneers/licklider.html. Accessed August 13,2007.Licklider, J. C. R. “Man-Computer Symbiosis.” IRE Transactions onHuman Factors in Electronics, vol. HFE-1, March 4–11, 1960.Available online. URL: http://www.memex.org/licklider.pdf.Accessed August 13, 2007.Licklider, J. C. R., and Robert W. Taylor. “The Computer as aCommunication Device.” Science and Technology, April, 1968.Available online. URL: http://gatekeeper.dec.com/pub/DEC/SRC/publications/taylor/licklider-taylor.pdf. Accessed August13, 2007.Waldrop, M. Mitchell. The Dream Machine: J. C. Licklider and theRevolution That Made Computing Personal. New York: Viking,2001.linguistics and computingThe study of human language and advances in computerscience have been closely intertwined. The field of computationallinguistics uses computer systems to investigate thestructure of natural language. In turn, the area of naturallanguage processing involves the creation of software thatcan apply linguistic principles to process written or spokenhuman language (see natural language processing,language translation software, and speech recognitionand synthesis).As simple low-level instruction codes began to evolveinto complex high-level programming language, languagedesigners had to struggle to give precise, complete, andunambiguous definitions for the language’s structure. Thisis essential for language users to be confident that theirprograms will yield the desired results. It is also importantthat developers trying to implement a language on differenthardware platforms and operating systems have rigorouslanguage specifications so the compiler on the new systemwill produce programs equivalent to those on the systemwhere the language was first developed.When computer scientists turned to linguistics for helpin defining programming languages, they found the workof Noam Chomsky, perhaps the 20th century’s preeminentlinguist, to be particularly helpful. Chomsky developed aconcept of formal language in which grammar could bespecified as a series of rules built up a level at a time.For example, at the lowest level, there is an alphabet fromwhich recognized words are generated. Next there are rulesfor generating phrases (such as a noun phrase consisting ofa noun with optional adjectives and a verb phrase consistingof a verb with optional adverbs). In turn, phrases can becombined to form sentences.Because grammatical structures are created by applyingrules to strings of symbols (words), the result is called agenerative grammar. Chomsky sought to apply this conceptof a “transformational generative grammar” as a universalstructure applicable to all human languages. Meanwhile,computer scientists could use formal grammar rules todefine the valid statements in programming languages (seealso Backus-Naur form). This in turn allows a compilerparser to break down high-level language statements andconvert them into low-level instruction codes that can actuallybe executed by the CPU (see assembler and parsing).As new languages and more powerful hardware gavecomputers increased power to deal with complex systems,computer scientists (and artificial intelligence researchers inparticular) applied themselves to the problem of computerprocessing of human languages. Success in this field mightlead not only to computer systems that humans could communicatewith far more naturally, but also to automatic machinetranslation that could, for example, allow an English speakerand a Chinese speaker to communicate via e-mail.However, developers of natural language systems faceformidable challenges. Most fundamentally, while computersprocess symbols using a restrictive, deterministic procedurethat Chomsky classifies as finite state (see finite statemachine), human languages must be understood usingthe more complex transformational grammar. The languageprocessing system must therefore have rules that cancope with the often ambiguous structure of actual humanspeech. (For example, does the word fly in a given sentencemean an insect, a baseball batted high in the air, or perhapsa zippered opening in one’s trousers?)One way to limit the problem is to deal with a restrictedrealm of discourse. For example, a natural language “frontend” to a database might assume that all input nouns referto entities that exist in the database, such as employees,positions, salaries, and so on. It then becomes a matter oftranslating a query such as “How many employees in thehuman resources department make more than $50,000 ayear” into something like:find quantity (employee.department = “humanresources”) and (employee.salary > 50,000)Understanding unrestricted text such as that foundin newspaper stories is much more complex, since fewerassumptions can be made about the subject of the discourse.Here the AI concept of frames can prove useful. A frame isa sort of script that describes the elements of life’s commonevents or transactions. For example, suppose a newsstory begins “Joe X was arrested yesterday for the murder