Natural Language vs. GUI: which is better?
Charles Hope
I am designing a front end for a database search tool to be used by novices.
So far I have two ideas for the best way this is to be done.
The first way is to present the user with a blank text input box and let them
enter their requests in natural English. They can be shown a list of sample
requests so they can get a feel for the possibilities. Mistakes will be
dealt with interactively as the sentence is parsed.
The second way is to use what I call a dynamic fill-in. The phrase "I am
searching for a " appears on the screen, followed by a pull-down menu
containing all the possible choices. When they select one, the rest of the
sentence appears, with appropriate text boxes and pull-downs. For instance
if they select "mashed potatoes" from the initial pull-down, the system
completes the sentence with "seasoned with " and checkboxes for salt, pepper,
butter, etc.
The second way is easier to implement, and I think that people might
appreciate the hand-holding provided. Does such a system make people feel more
secure and confident, or would people prefer to use natural language phrases
whenever possible? (We are assuming that the parsing of the English will not
be too difficult, for the sake of discussion here.)
I'm looking for the perspective of the computer illiterates, assuming they
understand pull-downs and whatever gadgets I might employ. What are the
results of research comparing people's preferences and effectiveness with
well-designed synthetic interfaces vs. being able to use their native
languages?
Many thanks!
____________
Charles Hope PENTIUM II- Close enough for government work.
(Disclaimer: May not apply to IRS, intelligence agencies, defense department
or any bureau needing to do math. http://www.x86.org/secrets/Dan0411.html )
http://www.walrus.com/~zod
Margaret Tarbet
On 18 Jun 1997 00:29:39 GMT,
z...@walrus.com (Charles Hope) wrote:
>
> I am designing a front end for a database search tool to be used by novices.
A major characteristic of novices, of course, is that few of them
stay so. Most almost imperceptably make the trip - the one-way
trip - to expert.
> The first way is to present the user with a blank text input box and let them
> enter their requests in natural English. ...
> The second way is to use what I call a dynamic fill-in. The phrase "I am
> searching for a " appears on the screen, followed by a pull-down menu
> containing all the possible choices. ...
> The second way is easier to implement, and I think that people might
> appreciate the hand-holding provided. Does such a system make people feel more
> secure and confident, or would people prefer to use natural language phrases
> whenever possible?
My intuition is that everyone would prefer the hand-holding as
novices, but then quickly progress (perhaps after only two or
three trials, depending on the sophistication of the NL
algorithm) to preferring the free-form interface, particularly if
requests could successfully be couched in "telegraphic" style.
Folk hardly ever want to waste time when they have a goal in
view, so they'll take whichever path seems likeliest to minimise
their cost, whether that cost comes from waiting around while the
interface assembles its stuff for prompting, or from having to
backtrack because the interface doesn't understand common
synonyms and thus gets confused by truly free-form input. Of
course, some won't ever use it enough to feel confident about the
f-f interface, while others will switch after their first trial.
Same as how folk choose menus vs keyboard shortcuts, really.
So the "right" resolution to your problem is (as usual! :-):
offer both. But if you really _can't_ offer both, offer the one
that is more certain to allow people to succeed, namely the
handholding version[1]. The potential experts may then curse
you for slowing them down to a crawl, but at least you won't be
cursed for providing an unusable system.
=margaret
---
[1] As you intimate, if you can do a good job on the free-form
interface, putting a prompting wrapper around it would be
trivial. Doing a good limited-domain NL interface, on the other
hand, is still _far_ from trivial, the examples of Shrdlu et al.
notwithstanding.
..........................................................
Margaret Tarbet
Software Art & Architecture Incorporated
Post Box 390 209, Cambridge Massachusetts USA
net: tar...@swaa.com; vox: +1-617-438-8647, fax: +1-617-438-4574
Comprehensive, Experienced, and Creative
Human-Factors/Marketability Design and Engineering.
Local or Long Distance. Satisfaction Guaranteed.
..........................................................
If I am not for myself, who will be for me? But if I am
only for myself, what am I? ... And if not now, when?
-- Rabbi Hillel, called The Babylonian (ca. 60 BCE)
Spiros Triantafyllopoulos
Charles Hope <z...@walrus.com> wrote:
: The first way is to present the user with a blank text input box and let them
: enter their requests in natural English. They can be shown a list of sample
: dealt with interactively as the sentence is parsed.
The problem is that very few people got natural language right in
the first place, even in a limited or restricted domain. Even a
dozen years or so ago when I was doing my MS CS in Natural Language
Processing (RIP), long before GUI's were the trend, we could see
that NLP queries were not such a good idea after all... In a limited
domain, it's doable, but even then, you will run into expert vs. novice
user preferences, you'll have to account for different vocabularies,
and so on. Having spent a couple of years programming the thing,
lemme tell you, it ain't trivial.
Later discussion with NLP types revealed some even more entertaining
problems, namely, that novice users had no clue as to what information
was in the system to begin with (not serious in a small school type
project, but lethal in a large enterprise management information system).
Expert users also were confused, in particular when forming complex
queries which can break the best of NLP systems.
A good reference (probably old, but what the *&^*^*&&) that I remember
being handy when I did my thesis writeup was:
[Lehnert 78] Lehnert, Wendel G., The Process of Question Answering:
A Computer Simulation of Cognition, Lawrence Erlbaum Associates,
Publishers, New York City, NY, 1978.
: The second way is to use what I call a dynamic fill-in. The phrase "I am
: searching for a " appears on the screen, followed by a pull-down menu
: containing all the possible choices. When they select one, the rest of the
: sentence appears, with appropriate text boxes and pull-downs. For instance
: if they select "mashed potatoes" from the initial pull-down, the system
: completes the sentence with "seasoned with " and checkboxes for salt, pepper,
: butter, etc.
This makes more sense, mostly because people have grown more 'used' to such
things. The technology to implement them is available, and things can be
prototyped in a hurry.
: The second way is easier to implement, and I think that people might
: appreciate the hand-holding provided. Does such a system make people feel more
: secure and confident, or would people prefer to use natural language phrases
: whenever possible? (We are assuming that the parsing of the English will not
: be too difficult, for the sake of discussion here.)
Speaking from experience here :-), parsing English is not a 'not too difficult'
task. Even with modern compiler construction tools (yacc, lex) the code
ended up over 10K lines (original)... Not for the faint at heart :-).
: I'm looking for the perspective of the computer illiterates, assuming they
: understand pull-downs and whatever gadgets I might employ. What are the
: results of research comparing people's preferences and effectiveness with
: well-designed synthetic interfaces vs. being able to use their native
: languages?
The illiterates will have problems one way or the other. A text interface
(essentially a blank screen) can be *very* intimidating (try the old style
OAG on AOL :-) if you don't believe me). If you can use graphical elements,
especially stuff that the user can relate to, i.e bitmapped buttons of
familiar real-world things, it becomes much easier.
Spiros (former NLP type, current GUI type, future HCI type)
--
Spiros Triantafyllopoulos Email: c2...@eng.delcoelect.com
Delco Electronics, Kokomo, IN http://www.primenet.com/~strianta