In this project, you will load the TRIPS ontology and use selectional restrictions to assign the correct semantic roles to semantic templates.

For this project you should use python3 and you should not require any dependencies outside of the standard library.

In this assignment you will be reading the TRIPS ontology and then using it to perform some word sense disambiguation tasks.

The TRIPS ontology is provided as a JSON file for easy loading. code/ontology.py demonstrates how to load a JSON file. If you run the code from the root directory of this project, you should see this:

project1> python code/ontology.py
there are 2907 nodes in the ontology

This file contains the semantic types and their arguments. Each entry has the following fields:

{
	"name": "ACTIVITY",          // name of the type
	"parent": "EVENT-OF-ACTION", // the parent of the current type
	"children": [
		"DIETING",                 // list of children of the type
		"PHYSICAL-ACTIVITY",
		...
	],
	"wordnet": [		           // list of wordnet mappings to the type
		"project%1:04:00",
		"project%1:09:00",
		"activity%1:04:00"
	],
	"arguments": [                     // list of semantic arguments for the type
		{
			"role": "AGENT",         // role filled by the argument
			"restrictions": [        // role restriction, see below
				"REFERENTIAL-SEM"
			],
			"optionality": "ESSENTIAL",  // whether the role is required or not
			"implements": "AGENT"
		},
		...
	]
}

This file contains mappings from words to ontology types. Note the words are all in their base form. The full lexicon also contains morphology to allow us to match other forms of the word (eg "run" and "running").

Once you have the ontology loaded there is some basic functionality that you will need to implement. The first will be a function to determine if a type is a subtype of a second type.

A second key function implements an inheritance algorithm to produce a set of possible roles and selectional restrictions for a type. For example, given the type CONSUME we will find directly that it has an AFFECTED role restricted FOOD, WATER or MEDICATION, and has an inherited AGENT role restricted to be ORGANISM. With CONVERSING, we find directly the AGENT AND AGENT1 roles as one of SOCIAL-GROUP-ABSTR, AGENT, SOCIAL-GROUP, ROBOT, SUPERNATURAL-BEING, or ANIMAL, and inherits the FORMAL role among others.

You will be given a semantic template showing words (that you can look up in your ontology) and semantic roles relating them as in the following example:

(eat :AGENT bat :AFFECTED date)

Note that bat can be NONHUMAN-ANIMAL or a MANUFACTURED-OBJECT and date can be a FRUIT or a TIME-POINT. Your answer should specify the correct type for all the words in the pattern. So the answer for the example would be

(CONSUME :AGENT NONHUMAN-ANIMAL :AFFECTED FRUIT)

Note this was a simple case since eat has only one meaning in the ontology. A verb like run has four meanings, EXECUTE, FUNCTION, GOVERNING and MOVE-RAPIDLY. Sometimes the right sense can be determined by the roles that are given independent of the argument words. For instance, given

(run :AGENT he :AFFECTED machine)

only senses of run that allow both the AGENT and AFFECTED roles would be possible. In some cases, there are multiple solutions possible. In that case, your program should list each of the possibilities.

We will provide several examples for testing. You should each additionally provide 10 of your own templates with the correct answers (one correct answer for each, removing any output predicates which satisfy the restrictions but have the wrong meaning).

Submit a [yourname].zip file on blackboard containing the folder structure provided. Your code should be in the folder entitled code and your data should be in the folder entitled data. Make sure to implement the methods in code/test.py to make grading easier.

The submission should contain:

• The original files in their current locations
• Any additional code files you need for your system to work
• An implemented code/test.py file.
• report.{txt|pdf} containing a description of your work including any specific running instructions and an outline of the algorithms you have implemented
• input/[yourname]-templates.txt containing a list of at least 10 templates you have come up with
• [yourname]-templates.gold.txt containing the correct answers for the templates you came up with