Grammex: defining grammars by example

Parsers are powerful tools for computer understanding of text, whether the language is a natural language or a formal language. To make the computational power of these tools fully available to an end user, a parser should be userextensible. Until now, a user who wished to control a parser was forced to write or edit a grammar, a text file containing rules. Editing grammars is often difftcult and error-prone for end users since the effect of writing specific rules, and interaction between rules, can often be unclear. Grammex [“Grammars by Example”] is the first direct manipulation interface designed to allow ordinary users to define grammars interactively. Instead of writing a grammar in an abstract rule language, the user presents concrete examples of text that he or she would like the parser to recognize. The user describes the text by selecting substrings, and choosing possible interpretations of the text from popup menus of suggestions heuristically computed by Grammex. Grammex compiles grammar rules that can be used as the input to a traditional parser. PARSERS AND GRAMMARS Parsers are beginning to be deployed as an integral part d the text editing facilities available across all computer applications. Examples are Apple Data Detectors [4] and the Intel Selection Recognition Agent [5]. Apple Data Detectors finds phone numbers, email addresses and URLs in free text These facilities allow automatic recognition of simple, commonly occurring text patterns such as e-mail addresses, URLs, or date formats. Typically, the set of patterns recognized by the parser is to be programmed by a highly expert user, a grammar writer skilled in computational linguistics. The end user is merely expected to invoke the parser and use its results. However, no set of patterns supplied by experts can be