Cool Tree Of The Week

redraw-bug.ps

Oops. Look what happens if you keep switching autoredraw on and off... XCDG eventually gets confused about which way is up!

I have no idea how it is even possible that a node is drawn below the sentence. And no, I can't reproduce it.

Here is a detail view if you can't view the whole tree well.

huge.ps

Now we know why the European Constitution was voted down! Honestly, who wants their life, rights and pursuit of happiness to be regulated by an unreadable mess like this? Without Björn's work on selective redrawing, this tree would have taken a day to correct.

inversion.ps

Look, the constraint that requires three-part subjunctive auxiliary chains in subclauses to be inverted actually as a purpose! The correct reading would be "...wenn der Online-Dienst tatsächlich an diese Börse würde gehen wollen" (although to be honest, that doesn't sound very much better to me either). Anyway, even very obscure conditions do eventually get their day in the sun, and at barely two years of age, too! The father is so proud.

IE-s10.ps

Look, ma! All those patient seller and product annotations done by Maria Zerwakis have finally been integrated into proper dependency annotations. We may or may not do something interesting with these trees in the upcoming information extraction seminar. Note how the ancillary NIL dependencies translate into upwards arrows that seem to point right at the word in question (XCDG upgrade required).

foo.ps

Stefan Klatt has replied at last! Here is a representation of a NEGRA sentence with our syntax annotation, combined with the topological field model analysis that his program assigns to it. By using this additional information, he claims that general parsing can be improved significantly.

We shall see whether this is actually true in combination with WCDG. In theory, knowing the boundaries of topological fields will rule out lots of dependency edges and at least make the parsing problem smaller. In practice the effect will probably be much dimished by errors in the TFM predictions. More on that next week.

Warum soll ein Unternehmen nicht Waschmaschinen wollen?

stupid2.ps

"Sehr geehrte Herren Aktionäre, im folgenden Geschäftsjahr werden wir den Schwerpunkt unserer Aktivitäten auf die Akquisition von mehr Waschmaschinen legen."

Und warum sollten Kochherde nicht als Programmierer arbeiten (oder von Bildschirmen handeln)?

Lexikalisierung tut not.

When someone sends me an email with a 42-word sentence, I have the compulsive urge to parse it automatically... and here is the result:

foo-s1.ps

Note the almost perfect (41/42) syntactic accuracy. (Spot the misattachment!) Note the robustness against the misspelled verb. Note the correct detection of the rare vocative. Note the elegant use of inversion in the fourth verb group. Note the correct combination of all 5 subclauses, with its very-long-distance coordination.

All done with CDG + deutsch grammar and no additional tweaking whatsoever.

I am happy.

Why oh why did TnT have to classify `die' as a relative pronoun?

stupid1.ps

Now we get four wrong attachments just because of one lousy NP-embedded PP.