Skip navigation.
Home
Semantic Software Lab
Concordia University
Montréal, Canada

Mulitpax not working with Supple Parser - Reoccurred

Printer-friendly versionPrinter-friendly versionPDF versionPDF version

Hi,

I am having the same problem as described here:
http://www.semanticsoftware.info/forum/mulitpax-not-working-supple-parser

I have reproduced the problem on 2 computers (GATE 6.0 win 7 64-bit, GATE 5.2.1 WinXP 32-bit). I have tried both Prolog Cafe and SWIProlog, the same problem occurs. I get no annotations under the annotation set category "MultiPaX", only a grey arrow.

Could someone provide me with their parameters for these processes, or send an .gapp application for me to try.

Also, where can I see an evaluation of the performance of all the parsers rasp3, Minipar, stanford, supple, etc..?

I would really appreciate any help here, thanks!

DETAILS:

The GATE message console displays:

---

Executing SWI prolog with the command line:

plcon.exe
-x
C:\Program Files\GATE-6.0\plugins\Parser_SUPPLE\supple.swi
--
-o
C:\Users\edan\AppData\Local\Temp\SUPPLE--OUT--8398375940932563999
C:\Users\edan\AppData\Local\Temp\SUPPLE--IN--6806409787049689737

extracting supple PAS!

---

My PRs are:

Document Reset
ANNIE English Tokeniser
ANNIE Sentence Splitter
ANNIE POS Tagger
Morphological Analyser
SUPPLE Parser
MultiPaX

All processes are set to default (blank parameters), except:

ANNIE POS Tagger:
baseSentenceAnnotationType - Sentence
base TokenAnnotationType - Token
outputAnnotationType - Token

Morphological Analyser:
considerPOSTag: true

SUPPLE Parser:
semanticsSetName: supple

MultiPaX:
outputASName: MultiPaX
parserASName: supple

Under the default annotation set I get:
Sentence
SpaceToken
SyntaxTreeNode
Token
parse

Under supple:
semantics

MultiPaX (greyed out)

The document string:
The quick brown fox jumps over the lazy dog

OUTPUT from supple:

semantics supple 0 44 2940 {qlf=[fox(e2), number(e2, sing), adj(e2, quick), adj(e2, brown), det(e2, 'The'), realisation(e2, offsets(0, 19)), jump(e1), time(e1, past), aspect(e1, simple), voice(e1, active), over(e1, e3), dog(e3), number(e3, sing), adj(e3, lazy), det(e3, the), realisation(e3, offsets(32, 44)), realisation(e1, offsets(20, 44)), realisation(e1, offsets(20, 44)), lsubj(e1, e2)]}

rene's picture

nothing to extract here...

As you can see from the two "semantics" annotations, SUPPLE was not able to create a single (complete) parse tree for this sentence. In particular, none of the two partial parse trees that were generated contain a predicate-argument structure, so there is nothing for MultiPAX to extract, hence the MultiPAX output annotation set is empty. If you try MiniPar on this sentence, you will see that MultiPAX can extract the triple (fox, jumps, dog). Note that it is one of the design ideas behind SUPPLE to only generate trees where it is sure the tree is correct, unlike most other parsers, which typically give you their best guess. Hence, with SUPPLE you will generally get lower recall but higher precision.

Regarding a performance comparison between the various parsers, have a look at our paper: Krestel, R., R. Witte, and S. Bergler, "Predicate-Argument EXtractor (PAX)", New Challenges for NLP Frameworks, Valletta, Malta : ELRA, pp. 51--54, May 22, 2010. If you want to know how correct each parser is by itself, you'll have to look into the original publications for each system.