interactive keys

[Mike Dallwitz]

Rudy Jocqué wrote:

> Apparently nobody mentioned X:ID. Does someone have experience with that
> package?

This program is free, and is available at
http://uio.mbl.edu/services/key.html. The current version is shown on the
main page as 0.66 (July 2004). I don't think it has been updated since then,
although when you click on a link to a key you get the message 'You are
looking at an old version of X:ID; in 1 second you will be automatically
re-directed to the current version'. The link to the X:ID forum, which might
have clarified this, doesn't work.

The system lacks some important features, such as error tolerance and
character reliabilities, and the 'Best characters' algorithm is worse than
useless (see below).

An interesting feature is the ability to change the appearance of the user
interface by means of style sheets. However, the examples currently
available are unsatisfactory in various ways. The style sheet that emulates
the Lucid2 interface is probably the best, except for a serious bug -  the
'feature' description is not shown in the 'Character States Chosen' pane.
This often results in meaningless lines such as 'triangular' instead of
'Stem shape in cross section triangular'.

At the request of the authors, I made some comments on the program in May
2005. These are reproduced below, except for some rather detailed ones on
the user interfaces. I tested only the player, not the builder.

Note that these comments don't constitute a review of the program, which has
many good features. I mentioned only the faults, so that they could be fixed.

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

'Misinterpretation'

The default setting should be 'Allow uncertainties and misinterpretation',
and it should be possible to change this while using a key.

Bingo

The Lucid 'Bingo' option is harmful in the context of identification - see
'Principles of Interactive Keys' at
http://delta-intkey.com/www/interactivekeys.htm under the heading
'Differentiating attributes'.

Best

The X:ID 'Best' is a extension of 'Bingo', and isn't suitable for finding
the best characters for use in an identification.

The problem with 'Bingo' and the X:ID implementation of 'Best' (which is
also used in MEKA) is not just that they are unhelpful for identification -
they are actually harmful. You would do better by selecting characters at
random, provided you used the 'Prune Redundants' option (if it worked
correctly, which it doesn't in X:ID).

For example, in the 'Aquatic Plants' key, suppose that your specimen is
Alternanthera. In the 'Best' list, the first character state that is
exhibited by Alternanthera is 'Leaf arrangement on stem (phyllotaxis I):
opposite distichous 17'. This is 142nd in the list. Thus, you have to
examine and reject 141 character states before arriving at the correct one,
and the probability of doing this without error is almost zero. If you do
manage to do this correctly, you still have 17 taxa remaining. In the new
'Best' list, the first character state exhibited by Alternanthera (Leaf
venation: pinnate 4) is 87th on the list, and 4 taxa remain. And so on.

By comparison, using the 'Best' character found by Intkey at each step, and
selecting the first state exhibited by Alternanthera, results in an
identification after 5 characters have been used.

Prune Redundants

I don't see the point of this option as it is currently implemented in X:ID.
Also, it's confusing because it puts into 'Character states chosen' states
that are not necessarily exhibited by the specimen.

As implemented in Lucid, the option has two effects, which I describe in
'Principles of Interactive Keys' under the headings 'Removing redundant
characters' and 'Removing redundant character states'. The former effect is
useful, but the latter is harmful. In Intkey, redundant characters are
automatically removed in the 'Best' calculation. I think that having it
available separately is an unnecessary complication for most users.

Similar Taxa

The results given by X:ID are substantially different from those given by
Lucid. I haven't investigated the reason for this.

Similarities and Differences

The output from this option is almost unintelligible. For comparison, I've
attached the differences between the genera of Cyperaceae, as output by
Intkey (diff_intkey.rtf) and X:ID (diff_xid.txt). [These files are not
attached to this posting.] The X:ID output is seemingly in almost random
order (even the states of a given character are not necessarily contiguous),
and there is no indication of which taxa exhibit the states. In the Intkey
output, notice the bold parts; these indicate characters for which at least
two of the taxa have non-overlapping states.

Numeric Characters

I gather from the forum that X:ID can handle numeric characters, but I
couldn't find any examples in the keys at the Web site.

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

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