> The patent for the tools that create Ergo Linguistic Technologiess’
> software has just been published by the U.S. Patent Office. Copies
***Incredibly long snip***
This is my first post to this group. I hope this isn't out of line, but I
was too tempted by the claims of this post to resist checking the online
demo. I tested it with three different problems:
1. Can it distinguish between the monotransitive and ditransitive senses
of verbs that take both forms.
Example: John served dinner.
John served Mike dinner.
John served Mike pancakes and sausages.
The parser suceeded in the first two sentences. It failed in the third
sentence, returning that 'Mike pancakes and sausages' was a direct object.
2. Prepositional attachment to verb phrases
Example
I ate spaghetti with wine
I ate spaghetti with a fork
I ate with a fork spaghetti
The parser suceeded in the first case but failed to associate the
preposition with the verb in the second. In the third it returned that 'a
fork spaghetti' was the object of ate.
3. Split verbs
Example
I need to drop off my clothing at the cleaners.
The parser returned I as the subject of the verb "need drop",
"to drop off my clothing at the cleaners" as the direct object of the verb
"need drop", and "my clothing at the cleaners" as the object of the
preposition "off".
I looked up the chimney
The parser correctly parsed this sentence
I looked up the phone number
The parser could not identify 'look up' as the verb. The parse was the
same as above.
To be fair, I tested it in its example domain, airline questions:
What is the take off time for the New York flight?
The parser returned "the take" as the subject of the verb "is" and
"off time for the New York flight" as an undefined object of the verb "is"
A couple of other interesting failures to try:
Spill soda on the floor and you will be in trouble.
We will fight in the forest, in the valleys, in the hills.
I know not what she does.
To give is better than to receive
I live in New Jersey and commute to work
I need to fly to New York and then to Boston
What flights from Memphis are there to Charlotte
I am a neophyte in NLP so I chose pretty simple parsing challenges.
These are the problems which I drew from an undergraduate level class on
the topic. I was fortunate enough to sit in on a demo from IBM on work
they're doing with probabilistic phrase grammars which could handle many
of these types of examples. My unprofessional opinion is that this
grammar has the same problems that every other basic grammar has.
I'm curious if others think the challenges I posed were reasonable.
Perhaps someone else can find a class of sentences that this system does
parse well.
Thank you,
David Campbell
Sunflower Sonnet Number One
But if I tell you how my heart swings wide
enough to motivate flirtations with the trees
or how the happiness of passion freaks inside
me, will you then believe the faithful, yearning freeze
on random, fast explosions that I place
upon my lust? Or must I say the streets are bare
unless it is your door I face
unless they are your eyes that, rare
as tulips on a cold night, trick my mind
to oranges and yellow flames around a seed
as deep as anyone may find
in magic? What do you need?
I'll give you that, I hope, and more
But don't you be the one to choose me: poor.
June Jordan