[Attempto] APE bug?

[Kaarel Kaljurand]

Hi,

On Mon, Jun 20, 2011 at 04:35, Kenneth Jones <kennethjone at gmail.com> wrote:
> When I preceded the word "children" with the definite determiner, giving:
> The a:textual n:serialization of an a:XML element is the concatenation of
> "<" and the name of the element and the a:textual n:serializations of the
> attributes of the element and ">" and the a:textual n:serialization of the
> sequence of the children of the element and "</" and the name of the element
> and ">".
> the error message changed to "time_limit_exceeded". The amount of time spent
> by APE was:
> overall: 20.7 sec (tokenizer: 0.010 parser: 0.020 refres: 16.420)

on my laptop, this sentence fails to parse after 14 seconds with an
error message from the reference resolution module (refres): Noun
phrase conjunction refers anaphorically to itself.
This message is repeated in the output 43 times, so I would call this
a bug (i.e. there is some unwanted backtracking somewhere).

Some observations:

* If I remove the adjectives "a:textual" and "a:XML" (or hyphenate
them to the following noun) then runtime drops to 6 seconds. But
adding them to the user lexicon does not help. It seems strange that
adjectives affect performance like this.

* Consider these two snippets:

(1) There is a man. A group is the man and a dog of the man.
(2) There is a man. A group is a dog of the man and the man.

Snippet (1) is parsed correctly. Snippet (2) gives (correctly?) the
error message "Noun phrase conjunction refers anaphorically to
itself." The confusing thing here is that sentences (1) and (2) are
not semantically equivalent (as one would expect, for these sentences
at least): in (2) 'and' binds stronger. I don't know if it's possible
to come up with a better interpretation rule in this case but in any
case we should describe the interaction of 'and' and 'of' in the
interpretation rules doc.

To locate the reason for the error in the original sentence I applied
some massive simplification to the original sentence and arrived at:

(3) "Fido" is the name of the dog of a man and a woman.

Sentence (3) gives the above error message but I think it should not.

In general, I'm actually not sure how to express the original sentence
in ACE. You use the definite article and the of-construction with an
intention of building a functional relationship, e.g. by "the sequence
of the children" you mean a _single_ ordering of _all_ the children,
but in ACE it means "a sequence of at least 2 children".

But in order to get rid of NP-conjunction (because it behaves
confusingly in the context of 'of') but still be able to express
ordered items you could use lists ([..]) or the concatenation operator
& (which seems appropriate for your sentence anyway). The shortcoming
with these constructors though is that you cannot use full NPs as
arguments, i.e. you would have to map your NPs to variables first,
e.g. this would work:

(4) If
there is an element, and
there is a name N of the element, and
there is an attributes-serialization A of the element, and
there is a children-serialization C of the element
then
a serialization of the element is
"<" & N & A & ">" & C & "</" & N & ">".

--
kaarel