Microsoft should big improve Word grammar check

[Harold F. Schiffman]

SEATTLE POST-INTELLIGENCER
http://seattlepi.nwsource.com/business/217802_grammar28.asp

A Word to the unwise -- program's grammar check isn't so smart
Monday, March 28, 2005

By TODD BISHOP
SEATTLE POST-INTELLIGENCER REPORTER

Microsoft the company should big improve Word grammar check.

No, your eyes aren't deceiving you. That sentence is a confusing jumble.
However, it is perfectly fine in the assessment of Microsoft Word's
built-in grammar checker, which detects no problem with the prose.

Sandeep Krishnamurthy thinks Microsoft can do a lot better.

The University of Washington associate professor has embarked on a one-man
mission to persuade the Redmond company to improve the grammar-checking
function in its popular word-processing program. Krishnamurthy is also
trying to raise public awareness of the issue.

"If you're a grad student turning in your term paper, and you think
grammar check has completely checked your paper, I have news for you -- it
really hasn't," he said.

Microsoft says it has been making continuous improvements in the
grammar-checking tool, and the company notes that the issue is more
complex than it might seem. Experts in natural-language processing say the
broader issue reflects a deep technological challenge beyond the current
capabilities of computer science.

"It is tremendously difficult," said Karen Jensen, a retired Microsoft
researcher who led the company's Natural Language Processing research
group as it developed the underlying technology for the grammar checker,
which debuted in 1997. "It gives you all kinds of respect for a human
being's native ability to learn and understand in natural language."

But Krishnamurthy, a professor of marketing and e-commerce at the UW's
Bothell campus, isn't convinced that the software giant is doing
everything it can -- and he supports his point with eye-catching examples.

He has crafted and posted for public download several documents containing
awful grammar. Depending on the version and settings, the Word grammar
checker sometimes detects a few of the problems. But it overlooks the
majority of them -- skipping misplaced apostrophes, singular-plural
inconsistencies, missing articles, sentence fragments, improper
capitalization and other problems.

An excerpt from one of his documents: "Marketing are bad for brand big and
small. You Know What I am Saying? It is no wondering that advertisings are
bad for company in America, Chicago and Germany. ... McDonald's and Coca
Cola are good brand. ... Gates do good marketing job in Microsoft."

With examples like that passing through unflagged, Krishnamurthy questions
whether Microsoft should even offer the grammar-checking feature in its
existing state.

"If you're including a feature in a widely used program like Microsoft
Word, it's got to pick up more things than it currently does," he said. "I
agree, the English language is very complicated, but I think we should
expect more from grammar check."

By comparison, the grammar checker in Corel Corp.'s WordPerfect Office 12
catches many of the errors in Krishnamurthy's test documents that aren't
detected by the Microsoft Word 2003 grammar checker, even set at the
highest sensitivity to errors.

In fact, there is room for Microsoft to make incremental improvements in
Word's grammar checker, said Christopher Manning, assistant professor of
linguistics and computer science at Stanford University.

For example, he said, the Word grammar checker could benefit from greater
use of advanced probabilistic and statistical methods to analyze sentences
and flag problems. Microsoft has applied some of that more advanced
research to competitive and high-profile areas such as Web search and spam
detection.

Microsoft says the grammar-checker does use probabilistic techniques in
addition to more basic, rules-based methods. But with further use of
advanced approaches, it appears possible for Word's grammar checker to
improve, Manning said. However, he said, "It still wouldn't be as good as
a good human editor."

Microsoft calls that the fundamental issue. Responding to an inquiry about
Krishnamurthy's examples, the Microsoft Office group said in a statement
that the grammar checker "was created to be a guide and a tool, not a
perfect proofreader." Microsoft also makes that point in Word's product
documentation.

The statement added, "It is possible to list a number of sentences that
you would expect the Word grammar checker to catch that it doesn't. But
that doesn't represent real-world usage. The Word grammar checker is
designed to catch the kinds of errors that ordinary users make in normal
writing situations."

It would be possible to "dial up the sensitivity" of the Word grammar
checker to catch more errors, the company said. However, that could also
cause it to flag sentences considered correct in colloquial usage.

That would risk making the tool more intrusive than people want, the
company said. In fact, Microsoft dialed down the sensitivity of the
grammar checker in certain respects starting in 2002, responding to
customer feedback. For example, some people objected when the tool flagged
sentences of more than 40 words as "perhaps excessively complex."

Krishnamurthy said he considers the company's view too simplistic. He
suggested that Microsoft further increase the available settings, beyond
the current options, to let people essentially "pick the level of
intrusion." He also said the company should offer an add-on for people who
need extra help, such as students for whom English is a second language.

As it now stands, the tool helps good writers but "really doesn't help bad
writers at all," he said.

Krishnamurthy, 37, grew up in Hyderabad, India. A textbook author and a
frequent contributor to scholarly journals, he is passionate about writing
and the English language.

But how did a marketing and e-commerce professor become a grammar-checking
crusader? While always stressing the importance of writing well in the
first place, Krishnamurthy would also routinely tell his students to run
the Word spelling and grammar checks as a precaution before turning in
their papers.

Then, last year, one student turned in a badly written report.

"The least you could have done is run spell-check and grammar-check,"
Krishnamurthy said.

"But I did!" the student said.

That prompted the professor to investigate, and he began discovering blind
spots in the Word grammar-checking tool. Krishnamurthy ultimately decided
to assemble specific examples of bad grammar that made it through
undetected. He began circulating them last week via e-mail to friends,
colleagues and Seattle-area media. He also created a Web page for the
purpose: http://faculty.washington.edu/sandeep/check.

The professor is careful to point out that he's not out to bash Microsoft.
But he says the company is spending too much energy on extraneous
capabilities, while neglecting core features such as the grammar checker.
Among other things, Microsoft is trying to expand the market for Microsoft
Office by adding a series of related server-based programs.

Office and related software make up Microsoft's second-most profitable
division, bringing in more than $7.1 billion in operating profit in the
last fiscal year. The core Office programs dominate the market.

Despite the lack of intense competition, there is a business incentive for
Microsoft to invest in core features, said analyst Rob Helm, research
director at Kirkland-based research firm Directions on Microsoft. That's
because one of the company's biggest challenges is persuading customers to
upgrade from older versions of its own programs.

By making improvements to features such as the grammar and spelling
checkers, Microsoft "can give people an additional incentive" to shift to
the newer version, Helm said.

Jensen, the retired Microsoft researcher who worked on the original
grammar-checking technology, said major advances would involve making
computers understand sentences in ways that humans would.

As an example, she cited one of the sentences used in Krishnamurthy's
sample documents: "Gates do good marketing job in Microsoft." Only by
knowing that "Gates" probably refers to Bill Gates -- and not to the
plural of the movable portion of a fence -- would the program know to
suggest using "does" instead.

"It's this level of understanding that you just can't expect a computer to
have at this point," Jensen said. "Someday, of course, it would be great,
but we're not there yet."

In the meantime, Krishnamurthy is spreading the message. He doesn't
suggest that anyone stop using the grammar-checking tool, but he wants
people to fully understand its limitations and not consider it a
substitute for good writing and editing.

In one part of his Web site, he has posted a cautionary list of "top
writing mistakes" made by his students. No. 11: "Assuming that Microsoft
Word's spelling and grammar check will solve all writing problems."



--------------------------------------------------------------------------------

On the Net: faculty.washington.edu/sandeep/check

P-I reporter Todd Bishop can be reached at 206-448-8221or
toddbishop at seattlepi.com

P-I senior online producer Brian Chin contributed to this report.

 1998-2005 Seattle Post-Intelligencer