SEMA4J

SEMA4J is a Java wrapper for easily accessing FrameNet semantic parsing by the SEMAFOR parser. Since this parser is primarly designed as a command line tool, this wrapper gives you easy programmatic access to FrameNet semantic parsing. It is also very easy to configure and get started with, let's check it out.

Installation

In order to be able to execute SEMA4J, you additionally need to add the original files of SEMAFOR to SEMA4J to the resources folder. The files you need to add can be found on the corresponding GitHub repository (SEMAFOR Semantic Parser). Just download the repository there, and include to the orignal repository file structure, the stackedParserServer and models folders to the root. A more detailed description of that can be found on their GitHub repository. Finally, add these files to this program under main/resources/semafor. SEMA4J will call this implementation of semantic parsing and return the results programatically. Make sure to start the server in order to fasten up the annotation process later.

Note that the SEMAFOR tool seems to work only with a linux system currently. All tests were done with Ubuntu.

Example

First of all, create a FrameNetAnnotator instance by specifying your JAVA_HOME installation path.

FrameNetAnnotator fnAnno = new FrameNetAnnotator(javaHomePath);

Instantiating the FrameNetAnnotator that way sets up standard options for the parser. If you want to set specific options, create a FrameNetOptions object and then create the annotator.

FrameNetOptions fnOpts = new FrameNetOptions(true, true, false, FrameNetOptions.DECODING_TYPE_BEAM,
                                              FrameNetOptions.NO_GOLD_FILES, javaHomePath);
FrameNetAnnotator fnAnno = new FrameNetAnnotator(fnOpts);

Once you have created the annotator, you can start with the semantic parsing of strings. This can be achieved, by simply calling the corresponding fetchFNResult methods as shown below. This method returns a list with all frames invoked by the assigned string. Afterwards you can access all properties of a frame and also iterate through the frame elements.

List<Frame> frames = fnAnno.fetchFNResult("Send you application to the university");
for(Frame frame : frames) {
//Do something with the frame
  for(FElement felement : frame.iterator()) {
    //Do something with frame element
  }
}

Often you want to collect multiple sentences to be annotated at once. Simply add them to the cache. From the results map, you can easily access to Frame list for each annotated string.

List<String> sentences = new ArrayList<>();
sentences.add("Wait for the results");
sentences.add("Receive acknowledgement from the university")
//Add to cache
fnAnno.addToCache(sentences);
//Get annotation results
Map<String, List<Frame>> frameAnnos = fnAnno.fetchFNResultsFromCache(sentence);
//Access frames for "Wait for the results"
List<Frame> frames = frameAnnos.get(sentences.get(0));