Nonsensical relation phrases with some remote arg2s.

[niranjanb]

Ollie produces some incorrect extractions with remote arg2s.

This seems to happen mostly when the remote arg2 is a date or a temporal expression and the relation phrase is based on a "be" verb.

Examples: The second extractions are all incorrect.

John Doe was in Washington on Thursday .

ollie extractions

0.91    John Doe    was in  Washington
0.88    John Doe    was on  Thursday

John Doe was in Washington for his vacation .

ollie extractions
0.91 John Doe was in Washington
0.85 John Doe was for his vacation

John Doe is in Washington to file his candidacy.

0.87    John Doe    is in   Washington
0.62    John Doe    is to file  his candidacy

[schmmd]

We want to fix (located as in situated):

He was located in Washington on Monday.

But it's problematic because we don't want to lose (located as in found):

The girl was lost. She was located deep in the woods on Monday.

[schmmd]

Maybe the "on Monday" arg2 is only useful if we have something else with the same relation or we can't have anything else.

She was located on Monday.
He filed a lawsuit on Monday.

But we're chilling out on this issue and everyone thinks both are groovy.

[mishumausam]

One option might be to disallow communication verb followed by a preposition.

"Microsoft returns high dividends", Steve Balmer said on Thursday.

[mishumausam]

Another option might be to add the first PP in the rel phrase (if there is no dobj and a secondary PP is in the arg2).

[schmmd]

Disallowing communication verbs followed by prepositions eliminates good extractions.

Obama spoke to the Jewish lobby.

[schmmd]

We have a few categories. Nonsensical, uninformative, and good. I will categorize our examples so far with N, U, or G.

John Doe was in Washington on Thursday .

G    0.91    John Doe    was in  Washington
N    0.88    John Doe    was on  Thursday

John Doe was in Washington for his vacation .

G    0.91 John Doe was in Washington
N    0.85 John Doe was for his vacation

John Doe is in Washington to file his candidacy.

G    0.87    John Doe    is in   Washington
N?    0.62    John Doe    is to file  his candidacy

[mishumausam]

I will be happy with the extractions:

John Doe, was in, Washington

John Doe was in Washington on, Thursday

John Doe, was in Washington for, his vacation

John Doe, is in Washington to file, his candidacy

So, the policy that we discussed should work.

Mausam

From: Michael Schmitz [mailto:[email protected]]
Sent: Wednesday, January 09, 2013 12:00 PM
To: knowitall/ollie
Cc: mishumausam
Subject: Re: [ollie] Bad relation phrases with some remote arg2s. (#11)

We have a few categories. Nonsensical, uninformative, and good. I will categorize our examples so far with N, U, or G.

John Doe was in Washington on Thursday .
N 0.91 John Doe was in Washington
G 0.88 John Doe was on Thursday

John Doe was in Washington for his vacation .
G 0.91 John Doe was in Washington
N 0.85 John Doe was for his vacation

John Doe is in Washington to file his candidacy.
G 0.87 John Doe is in Washington
N? 0.62 John Doe is to file his candidacy

—
Reply to this email directly or view it on GitHub #11 (comment) .

Image removed by sender.