Palm pilots - possibly a better usage & interactive keys

[Mike Dallwitz]

Stinger Guala wrote:

> As exceptionally good as many interactive key programs are right now, I
> don't think that the community will truly embrace them until they run
> ubiquitously and without plugins.

It's worth noting that Java-based keys such as Navikey and Lucid3, although
running on several platforms, do require the Java Runtime Environment to be
installed. This is a fairly large download: 15MB, compared with 2MB for Intkey.

> The past, in my opinion, has been the MS Office try-to-be-everything-to-
> everyone-approach but the future will look more like Google. I think in the
> very near future we'll have lightweight portable keys that go everywhere and
> really really smart server based keys for either online ID or for
> downloading those lightweight portable keys. For example, there are no
> "Best" characters.

'Best characters' should not be an optional extra - it's essential for
accurate identification for most users. (The exceptions are users who are
very familiar with the group, but even these users can benefit from using
it, especially in the later stages of an identification.) Without this
feature, it would not be surprising to find interactive keys performing
worse than conventional keys.

Although Intkey has many options, it operates by default in 'Normal' mode,
which contains only a few simple options, most of which are essential for
effective identification. Most users need never look at 'Advanced' mode
(which, however, is very much simpler than, say, MS-Word or Adobe
Photoshop). It's similar in this regard to Google - though we thought of it
first :-).

In Normal mode, the menu system is very small, containing only the 8 options
'File > New Data Set / Advanced Mode / Exit', 'Window > Tile / Cascade /
Close All', 'Help > Introduction / About'.

Identification is operated entirely via the buttons, of which there are 11:

     Restart identification
     'Best' order [of the characters] *
     Natural order [of the characters] *
     Differences between the specimen and the remaining or selected taxa
     Set error tolerance
     Use a subset of the characters
     Find text in characters
     Information about taxa
     Differences between taxa
     Use a subset of the taxa *
     Find text in taxon names *

The ones marked * could perhaps be relegated to Advanced Mode, but they are
useful for information retrieval and unlikely to be harmful for
identification, except the one for 'Natural order' of the characters. (If
the 'Natural order' button were omitted, the 'Best order' button would be
omitted too, as this is the default.) Notice also the _absence_ of buttons
for accessing character images - this happens automatically.

There are also some very important features behind the scenes, that have
little or no impact on the user interface, e.g. character reliabilities
(weights), direct use of numeric values, and character dependencies.

Without these options and features, the identification process won't be as
quick, accurate, and robust as it could be.

The utility of most of these options is probably obvious, and some are
discussed in 'Principles of interactive keys'
(http://delta-intkey.com/www/interactivekeys.htm). I'd be happy to further
discuss the need for these options, but DELTA-L
(https://listserv.surfnet.nl/archives/delta-l.html) might be a more suitable
forum for this.

I can't see how a Google-like approach is applicable in this area, or in
other specialized areas such as word processing or image editing. Also,
recent efforts of Google to mimic human/intelligent responses have
inevitably introduced the lack of precision typical of humans, and this can
be a disadvantage for some purposes. For example, in Google you can no
longer find the pages that actually contain the terms you are searching for.

> There are a lot of optimization algorithms (divide the pile evenly, what the
> experts say is best, what other people use most, what the data are most
> complete in, the most easily observable etc.).

The Intkey 'best characters' algorithm combines all of these in a way that
has a sensible theoretical justification, i.e. minimizing the probability of
making an error (see Dallwitz, M.J. 1974. A flexible computer program for
generating identification keys. Syst. Zool. 23: 50–7).

> In some web environments (e.g. AJAX - see Amazon.com) you can have the key
> suggest what is REALLY the best character state to use based on all of those
> - and the key's own experience with thousands of people in millions of
> keying sessions

You mean 'the best character', not 'the best character state'; knowing the
latter would be ideal, but it's some way down the track yet :-).

I think that the philosophy that the REALLY best way of doing anything is to
copy what thousands of other people have done is dubious.

I don't have the paper to hand, but I think the key described in
     Wilson, J.B., and Partridge, T.R. 1986. Interactive plant
     identification. Taxon 35, 1–12.
used information from previous identifications (by all users) to guide the
current one. Presumably, though, it relied on feedback about the accuracy of
previous identifications, and this would be difficult to obtain in an
uncontrolled environment such as the Internet.

> ... give you help when you actually need it, not just when you think you do.
> It isn't as scary as it sounds and it isn't far off.

It's certainly scary, and I hope it's a long way off! In general, the
computer would have to read your mind, and probably know the correct answer
as well. (Of course, the latter wouldn't be a bad thing, and will presumably
happen eventually. It would make all this discussion pointless.)

> DeLTa was the standard for interactive keys for a long time. Nexus is a more
> broadly accepted standard for this kind of matrix ...

Nexus is not really suitable for this purpose - see 'A Comparison of Formats
for Descriptive Data' (http://delta-intkey.com/www/compdata.htm).

--
Mike Dallwitz
Contact information: http://delta-intkey.com/contact/dallwitz.htm
DELTA home page: http://delta-intkey.com