1(4), August 1984, page 1

What is EPISTLE?

Christine Neuwirth, David S. Kaufer, and Cheryl Geisler
Carnegie-Mellon University

Currently occupying several researchers at IBM's Thomas J. Watson Research Center in Yorktown Heights, New York, the EPISTLE project represents a long range attempt

to provide office workers, particularly middle-level managers, with a variety of application packages to help them interact with natural language texts.[1]

By developing a computational grammar of English and a 100,000 word on-line dictionary, the EPISTLE project team has been able to build a computer program, also called EPISTLE, that can perform sophisticated, comprehensive analyses of grammar and style for many types of English sentences--even sentence fragments! With its grammar and dictionary, EPISTLE can detect a variety of common grammatical mistakes, among them:

EPISTLE can also flag stylistic weaknesses such as jargon, non- preferred spellings, inappropriate slang, awkward phrasings, over-use of negatives or passives excessive distance between subject and verb, and so on.

EPISTLE from the Writer's Point of View

A writer who runs a file (containing some letter, manuscript or other document) through the EPISTLE program will see the document displayed on the terminal screen and, in the upper right hand corner, will see three small rectangles labeled "spelling," "grammar," and "style." The rectangles represent the types of textual critiques available. By using a light-pen or the terminal keyboard, the writer can choose which of these critiques EPISTLE will apply to the document. Suppose the writer indicates a choice for "grammar." The rectangle labeled "grammar" will turn red, and EPISTLE will color-code in red all the places in the document where it has discovered a grammatical error.

By moving the cursor and selecting among EPISTLE's on-line instructional options, the writer can examine each error in more detail. Suppose, for example, that EPISTLE has found "has wrote" in the document and marked it in red. The writer can move the cursor to the location of the error, then select to see EPISTLE's suggestion for a correction. EPISTLE displays "has written" above the error. Then the writer can select to see EPISTLE's explanation of the correction--in this case, a statement about the declension of the verb write. If the writer desires a more detailed explanation, he or she can access a third level of detail--a tutorial on the grammatical construct in question. At any point, the writer can either decide to take EPISTLE's advice, make his or her own correction, or leave the text as it is. EPISTLE thus helps the writer by locating errors, suggesting corrections, and giving explanations at the desired level of detail.

With each local correction, the red-marking for that error disappears. When the writer finishes all the corrections, the "grammar" box at the top of the file changes colors--from red to green! The writer can then proceed through spelling or style corrections in a similar way.


How can EPISTLE best be used by students and faculty in our colleges and universities? In 1984, researchers at IBM and Carnegie-Mellon will be working together to answer this question. IBM will be lending a copy of EPISTLE and providing technical support to Carnegie-Mellon's Writing Center. Faculty and graduate research assistants in the Center will work to assess (1) EPISTLE's capabilities, (2) its effects on the revision process, and (3) its potential as a teaching tool. In addressing these issues, we will look at three different types of users: professional writers, college students, and non-native speakers of English.

1. Capabilities

COMPUTERS and COMPOSITION 1(4), August 1984, page 2

To what extent do beginning and experienced college-level writers make the kinds of mistakes that EPISTLE detects and corrects? Can EPISTLE handle mistakes made by non-native writers? Are there frequently occurring errors that the system doesn't handle? If so, can the system be extended to include these? To answer these questions, we have begun to collect and analyze the papers that students write in freshman, upper-level, and ESL writing courses. We will use this corpus to explore the capabilities and potential of the program.

2. Effects on Revision

We expect that EPISTLE is bound to reduce the number of errors at the sentence level that writers have in a final draft. We would like to know whether these improvements occur because EPISTLE directs writers to attend to error, because it advises writers on how to correct error, or because it explains error. In other words, does EPISTLE simply remind people to do what they already know how to do, or does it enable them to do what they cannot or would not do for themselves? Are the advice and explanations appropriate for different users? Obviously, the answer to these questions may depend upon the experience of the writer.

We also expect EPISTLE to make the revision process more efficient. But will writers use this "saved time" to attend to more substantial, higher-level revisions or will they simply spend less time revising? Research suggests that beginning writers seldom revise discourse structures higher than the sentence. Will EPISTLE reinforce this pattern or give them the time to break it?

3. Potential as a Teaching Tool

Will extended use of EPISTLE change the pattern of errors in writers' first drafts? Will writers come to understand and avoid their chronic problems or will they rely on EPISTLE to "catch" them? How will they fare when EPISTLE is no longer available to them--will they have learned from the program or will they revert to their original patterns? These questions seem particularly important to ask as we think of giving a tool which was developed for use by professionals to beginning writers.

How should the program be configured to optimize learning? Currently the program detects all grammatical errors in a text in the order of their occurrence. Will students be overwhelmed by the number and variety of errors? Would modifying the program to detect a subclass of errors at a time facilitate learning? Currently the program gives the student the exact location of the error and the correction. Would it be better to modify EPISTLE's instructional options so that students must find and correct the problems themselves--with EPISTLE simply checking the students' work?

Can EPISTLE be used to help diagnose writing problems? Can it be used both to help identify those problems a writer has and rule out those he or she doesn't have? Can it be useful in diagnosing problems in omission as well as commission, e.g., can it be used to identify syntactic constructs that a basic writer or ESL student doesn't use, but perhaps should? We plan to explore EPISTLE's potential as a diagnostic tool, and how best to use it in different learning environments.

Can EPISTLE be used effectively in different instructional paradigms? For example, can it be used to increase upper-level students' knowledge of the structure of language, helping them to understand and discover general principles?

We find this last question particularly appropriate and challenging. The construction of EPISTLE itself marks an attempt to find general principles of English grammar and style. Limitations in EPISTLE are in large part limitations in our own understanding of the general principles of language. Interacting with EPISTLE will, we hope, teach students an invaluable lesson--that principles of good writing are a matter of problem-solving, discovery, and hypothesis-testing. To teach students this lesson is our most important challenge in working with EPISTLE.

1. See G.E. Heidorn, K. Jensen, L.A. Miller, R.J. Byrd, and M.S. Chodorow, The EPISTLE test-critiquing system, IBM System Journal , 21 (1982), 305-325.