NAME

WordNet::Similarity::lesk - Perl module for computing semantic relatedness of word senses using gloss overlaps as described by Banerjee and Pedersen (2002) -- a method that adapts the Lesk approach to WordNet.

SYNOPSIS

  use WordNet::Similarity::lesk;

  use WordNet::QueryData;

  my $wn = WordNet::QueryData->new();

  my $lesk = WordNet::Similarity::lesk->new($wn);

  my $value = $lesk->getRelatedness("car#n#1", "bus#n#2");

  ($error, $errorString) = $lesk->getError();

  die "$errorString\n" if($error);

  print "car (sense 1) <-> bus (sense 2) = $value\n";

DESCRIPTION

Lesk (1985) proposed that the relatedness of two words is proportional to to the extent of overlaps of their dictionary definitions. Banerjee and Pedersen (2002) extended this notion to use WordNet as the dictionary for the word definitions. This notion was further extended to use the rich network of relationships between concepts present is WordNet. This adapted lesk measure has been implemented in this module.

Methods

$measure->initialize($file)

Overrides the initialize method in the parent class (GlossFinder.pm). This method essentially initializes the measure for use.

Parameters: $file -- configuration file.

Returns: none.

$lesk->traceOptions()

This method is internally called to determine the extra options specified by this measure (apart from the default options specified in the WordNet::Similarity base class).

Parameters: none.

Returns: none.

$lesk->getRelatedness

Computes the relatedness of two word senses using the Extended Gloss Overlaps algorithm.

Parameters: two word senses in "word#pos#sense" format.

Returns: Unless a problem occurs, the return value is the relatedness score, which is greater-than or equal-to 0. If an error occurs, then the error level is set to non-zero and an error string is created (see the description of getError()).

Usage

The semantic relatedness modules in this distribution are built as classes that define the following methods:

  new()
  getRelatedness()
  getError()
  getTraceString()

See the WordNet::Similarity(3) documentation for details of these methods.

Typical Usage Examples

To create an object of the lesk measure, we would have the following lines of code in the Perl program.

   use WordNet::Similarity::lesk;
   $measure = WordNet::Similarity::lesk->new($wn, '/home/sid/lesk.conf');

The reference of the initialized object is stored in the scalar variable '$measure'. '$wn' contains a WordNet::QueryData object that should have been created earlier in the program. The second parameter to the 'new' method is the path of the configuration file for the lesk measure. If the 'new' method is unable to create the object, '$measure' would be undefined. This, as well as any other error/warning may be tested.

   die "Unable to create object.\n" if(!defined $measure);
   ($err, $errString) = $measure->getError();
   die $errString."\n" if($err);

To find the semantic relatedness of the first sense of the noun 'car' and the second sense of the noun 'bus' using the measure, we would write the following piece of code:

   $relatedness = $measure->getRelatedness('car#n#1', 'bus#n#2');

To get traces for the above computation:

   print $measure->getTraceString();

However, traces must be enabled using configuration files. By default traces are turned off.

CONFIGURATION FILE

The behavior of the measures of semantic relatedness can be controlled by using configuration files. These configuration files specify how certain parameters are initialized within the object. A configuration file may be specified as a parameter during the creation of an object using the new method. The configuration files must follow a fixed format.

Every configuration file starts with the name of the module ON THE FIRST LINE of the file. For example, a configuration file for the lesk module will have on the first line 'WordNet::Similarity::lesk'. This is followed by the various parameters, each on a new line and having the form 'name::value'. The 'value' of a parameter is optional (in case of boolean parameters). In case 'value' is omitted, we would have just 'name::' on that line. Comments are supported in the configuration file. Anything following a '#' is ignored till the end of the line.

The module parses the configuration file and recognizes the following parameters:

trace

The value of this parameter specifies the level of tracing that should be employed for generating the traces. This value is an integer equal to 0, 1, or 2. If the value is omitted, then the default value, 0, is used. A value of 0 switches tracing off. A value of 1 or 2 switches tracing on. A value of 1 displays as traces only the gloss overlaps found. A value of 2 displays as traces all the text being compared.

cache

The value of this parameter specifies whether or not caching of the relatedness values should be performed. This value is an integer equal to 0 or 1. If the value is omitted, then the default value, 1, is used. A value of 0 switches caching 'off', and a value of 1 switches caching 'on'.

maxCacheSize

The value of this parameter indicates the size of the cache, used for storing the computed relatedness value. The specified value must be a non-negative integer. If the value is omitted, then the default value, 5,000, is used. Setting maxCacheSize to zero has the same effect as setting cache to zero, but setting cache to zero is likely to be more efficient. Caching and tracing at the same time can result in excessive memory usage because the trace strings are also cached. If you intend to perform a large number of relatedness queries, then you might want to turn tracing off.

relation

The value of this parameter is the path to a file that contains a list of WordNet relations. The path may be either an absolute path or a relative path.

The lesk measure combines glosses of synsets related to the target synsets by these relations and then searches for overlaps in these "super-glosses."

WARNING: the format of the relation file is different for the vector and lesk measures.

stop

The value of this parameter the path of a file containing a list of stop words that should be ignored in the glosses. The path may be either an absolute path or a relative path.

stem

The value of this parameter indicates whether or not stemming should be performed. The value must be an integer equal to 0 or 1. If the value is omitted, then the default value, 0, is used. A value of 1 switches 'on' stemming, and a value of 0 switches stemming 'off'. When stemming is enabled, all the words of the glosses are stemmed before their vectors are created for the vector measure or their overlaps are compared for the lesk measure.

normalize

The value of this parameter indicates whether or not normalization of scores is performed. The value must be an integer equal to 0 or 1. If the value is omitted, then the default value, 0, is assumed. A value of 1 switches 'on' normalizing of the score, and a value of 0 switches normalizing 'off'. When normalizing is enabled, the score obtained by counting the gloss overlaps is normalized by the size of the glosses. The details are described in Banerjee and Pedersen (2002).

RELATION FILE FORMAT

The relation file starts with the string "RelationFile" on the first line of the file. Following this, on each consecutive line, a relation is specified in the form --

func(func(func... (func)...))-func(func(func... (func)...)) [weight]

Where "func" can be any one of the following functions:

  hype() = Hypernym of
  hypo() = Hyponym of
  holo() = Holonym of
  mero() = Meronym of
  attr() = Attribute of
  also() = Also see
  sim() = Similar
  enta() = Entails
  caus() = Causes
  part() = Particle
  pert() = Pertainym of
  glos = gloss (without example)
  example = example (from the gloss)
  glosexample = gloss + example
  syns = synset of the concept

Each of these specifies a WordNet relation. And the outermost function in the nesting can only be one of glos, example, glosexample or syns. The set of functions to the left of the "-" are applied to the first word sense. The functions to the right of the "-" are applied to the second word sense. An optional weight can be specified to weigh the contribution of that relation in the overall score.

For example,

 glos(hype(hypo))-example(hype) 0.5

means that the gloss of the hypernym of the hyponym of the first synset is overlapped with the example of the hypernym of the second synset to get the lesk score. This score is weighted 0.5. If "glos", "example", "glosexample" or "syns" is not provided as the outermost function of the nesting, the measure assumes "glos" as the default.

So,

 glos(hypo(also))-glos(holo(attr))

and

 hypo(also)-holo(attr)

are treated the same by the measure.

AUTHORS

 Ted Pedersen, University of Minnesota Duluth
 tpederse at d.umn.edu

 Satanjeev Banerjee, Carnegie Mellon University, Pittsburgh
 banerjee+ at cs.cmu.edu

 Siddharth Patwardhan, University of Utah, Salt Lake City
 sidd at cs.utah.edu

BUGS

None.

To report bugs, go to http://groups.yahoo.com/group/wn-similarity/ or e-mail "tpederse at d.umn.edu".

COPYRIGHT AND LICENSE

This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 2 of the License, or (at your option) any later version.

This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.

You should have received a copy of the GNU General Public License along with this program; if not, write to

   The Free Software Foundation, Inc.,
   59 Temple Place - Suite 330,
   Boston, MA  02111-1307, USA.

Note: a copy of the GNU General Public License is available on the web at <http://www.gnu.org/licenses/gpl.txt> and is included in this distribution as GPL.txt.