
Format specification for SENSEVAL-3 results

The results of a system's processing should be reported in a single
file corresponding to the evaluation data file, with one line for each
test instance in the evaluation data.  The lines in this results file
do not have to be in any particular order.

Each line should include the following items, in the specified order:

1. reference id for the lexical item
2. a single space
3. reference number for test instance
4. a single space
5. list of sense tags with or without weights
6. an optional comment

In a lexical sample task, the reference id is from the 'lexelt' tag in
the XML data and the reference number is the 'instance_id'.  In an
all-words task, the reference id is 'text_id' and the reference number
is the 'head_id'.

If the list of sense tags is weighted, the weights (given as integers
or real numbers, and separated by a "/" from the sense tag) will be
normalised.  That is, they will be treated as probabilities.  If any
of the sense tags are missing weights, then none of the weights will
be used.  When there are no weights, a uniform distribution is used.

In case the results file contains two or more lines for the same
reference number for the same reference id, the first such line
will be counted as the system's answer and the subsequent lines will
be disregarded.

---------------------------------------------------------------------

Examples of well-formed lines:

brother.n 00001 501566
brother.n 00002 501566     999997     !!
brother.n 00006 501566/0.5 501573/0.4 503751/0.1
brother.n 00015 503751/94  999999/87             !! comment . . .

where 'brother.n' is an example of a lexelt field, '00001' is an example of
an instance_id, and '501566' is an example of a sense_tag.  Actual values
will depend on the data for the particular task.
---------------------------------------------------------------------

If you're into BNF, here's a precise definition:

lexical_sample_answer
   ::= lexelt instance_id sense_tag+ [!! comment]

all_words_answer
   ::= text_id head_id sense_tag+ [!! comment]

lexelt ::= lexical element (as defined in the lex-sample evaluation data)
instance_id ::= instance id (as defined in the lex-sample evaluaiton data)
text_id ::= text id (as defined in the all-words evaluation data)
head_id ::= head id (as defined in the all-words evaluation data)

sense_tag ::= senseid[/weight]
senseid ::= senseid from the lexicon for the task
weight ::= integer or positive real number
comment ::= text string

-----------------------------------------------------------------------
Note: Answer format adapted from the Senseval-2 scoring guidelines




