Interactive Vocabulary Acquisition in XTRA

This paper describes a practical solution to online dictionary update in X T R A , a machine translat ion system developed by X iuming Huang at the Computing Research Laboratory of New Mexico State University. The focus of the discussion in on IVES — an Interactive Vocabulary Enrichment System bui l t by this writer for X T R A , It reflects an ongoing effort at the laboratory to build embedded learning mechanisms in machine translation systems. Two types of learning are discussed, word learning and word sense learning. Each type of learning undergoes three routine processes: detection, acquisit ion, and evaluation. The emphasis of this paper is on the use of semantic preference violations in the detection of the need to learn new word senses. I I N T R O D U C T I O N Machine translation systems, bui l t using an AI approach, could be equipped with embedded learning mechanisms so that the systems would constantly update themselves and modify their behavior. Building such embedded systems should speed up the early realization of machine translat ion on microcomputers. This is because, except for the closed system of function words, an MT procram cannot be expected to have a 'complete' vocabulary at the time of installation. A n d it is almost impossible for the customer to purchase such systems that are adequate to start wi th. Learning mechanisms could be a remedy in this regard. It is usually easier to find a bilingual person than a software engineer to upgrade a system. The focus of the discussion is on IVES — an Interactive Vocabulary Enrichment System built by this writer for X T R A . Two types of learning are discussed, word learning and word sense learning. Each type of learning undergoes three routine processes: detection, acquisition, and evaluation. The emphasis of this paper is on the use of semantic preference violations in the detection of the need to learn new word senses. I I WORD L E A R N I N G According to a model of machine learning advocated by Cohen and Feigenbaum (1982), machine learning has four important elements: (a) the environment, (b) the learning element, (c) the knowledge base, and (d) the performance element. The environment supplies information to the learning element; the learning element uses this informat ion to make improvements in the knowledge base; the performance element uses the acquired informaUon to carry out its task; finally, feedback information gained during attempts to perform the task goes back to the learning element. In the context of word and word sense leaming in X T R A , the environment is the system's interaction with the user; the learning element is IVES; the knowledge base is the system dictionaries; and the performance element is the English sentence parser. Depending on the kind of information supplied by the environment to the learning element, there exist five different learning strategies. These are rote learning, learning by instruction, learning by deduction, learning by analogy, and learning by induction. (Michalski, Carbonell, and Mitchel l , 1986, p. 14). Word learning in X T R A belongs to rote learning. In rote learning the information supplied by the environment is directly accepted by the learning system. The strategy is elementary, not powerful enough to accomplish intelligent learning on its own, but it is an ' inherent and important part of any learning system' (Cohen and Feigenbaum, 1982, p. 335). The detection of the need to learn new words in X T R A is facilitated by X T R A ' s preprocessing routine. Whenever a certain word in the input sentence is found to be missing from X T R A ' s dictionaries, the preprocessing routine issues a warning and prints out the missing word. Whenever a word is detected as missing, IVES is called, and an interactive session with the system user begins. Dur ing the acquisition phase, information is elicited from the user to expand X T R A ' s dictionaries. IVES, as it stands now, handles four major categories of open system words, i.e., nouns, verbs, adjectives, and adverbs. A l l input into the system dictionaries is checked for its validity. Invalid entries are automatically rejected, and a second request for the same piece of information follows the rejection. Al though the default value of the target language is Chinese, IVES could handle any target language. When the acquisition phase is over, parsing continues f rom where it left off. Often, it succeeds as a result of dictionary expansion. However, it may also fail for reasons that wi l l be discussed in the next section. I l l WORD SENSE L E A R N I N G Word sense learning can be categorized as a special type of learning by instruction. It undergoes the same three processes as word learning does, i.e., detection, acquisition, and evaluation. The difference between word learning and word sense learning is that the detection process for the latter is much more complicated than that for the former. Also, in the course of the detection process,