Vulcan/DefinitionsExtractor
Overview
Definitions fall into a small number of syntactic patterns, each begins with “X is Y” giving a class that X belongs to, followed by some distinguishing characteristic. A typical pattern is “X is Y that rel Z”, e.g. “Carnivore is an animal that eats other animals.”. Another typical is “X is Y prep Z”, e.g.” Fossil is the remains of a living thing from many years ago”.
We need a much more aggressive extractor than Open IE 4.0 for definitions, since we want to include all words in a definition, and to create relations from prepositions and gerunds that Open IE doesn’t handle.
One approach is to create a Definition Pattern Matcher that takes as input the POS-tagged sentence, the parse tree, and the SRL-based tuples. It then matches the definition to one of the manually identified definition patterns and creates a conjuction of tuples.
Patterns
I analyzed a random set of 25 definition sentences and identified the following 15 definition patterns. I will continue this analysis on a larger sample.
- Patterns
1. X is Y that rel Z => (X, is-a, Y) & (Y, rel, Z) freq 3 Stamen is the part of the plant that holds the male cells for reproduction (stamen, is-a, part of the plant) & (part of the plant , holds, the male cells for reproduction) 2. X is Y that rel1 Z rel2 W => (X, is-a, Y) & (Y, rel1, (Z, re2l, W)) freq 2 Adaptation is a special trait that helps an organism survive in its environment (Adaptation, is-a, special trait) & (special trait, helps, (organism , survive in, its environment) 3. X is Y that rel1 when Z rel2 W => (X, is-a, Y) & (Y, rel1 when, (Z, rel2, W) freq 2 Volcano is a mountain that forms when red-hot melted rock flows through a crack onto earth's surface (Volcano, is-a, mountain) & (mountain, forms when, (red-hot melted rock, flows through, crack onto earth's surface) 4. X is Y rel by Z => (X, is-a, Y) & (Z, rel, Y) freq 1 Carbon dioxide is a gas breathed out by animals (Carbon dioxide, is-a, gas) & (animals, breathe out, gas) 5. X is Y rel Z => (X, is-a, Y) & (Y, rel, Z) freq 1 Experiment is an organized test designed to support or disprove a hypothesis (Experiment, is-a, organized test) & (organized test, designed to support or disprove, a hypothesis) 6. X is Y prep Z => (X, is-a, Y & (Y, is prep, Z) freq 6 Carbon dioxide is a gas in the air (Carbon dioxide, is-a, gas) & (gas, is in, the air) 7. X is Y prep Z rel W => (X, is-a, Y & (Y, is prep,( Z, rel, W)) freq 2 Inclined plane is a flat surface with one end higher than the other (Inclined plane, is-a, flat surface) & (flat surface, is with one end higher than, the other) 8. X is Y prep1 Z prep2 which W rel => (X, is-a, Y) & (Y, is prep1, Z) & (W, rel prep2, Z) freq 1 Stomata is the holes on the bottoms of leaves through which air and water pass (Stomata, is-a, holes) & (holes , is on, the bottoms of leaves) & (air and water, pass through, holes) 9. X is Y prep rel Z => (X, is-a, Y) & (Y, is prep rel, Z) freq 1 Classification is the grouping of things by using a set of rules (Classification, is-a, grouping of things) & (grouping of things , is by using, a set of rules) 10. X is having Y rel Z => (X, is having, (Y, rel, Z) freq 1 Hybrid is having two or more different things mixed together (Hybrid, is having, ( two or more different things, mixed, together) 11. X is Y; for example Z => (X, is-a, Y) & (Z, is example of, X) freq 1 Star is a huge, burning sphere of gases; for example, the sun (Star, is-a, huge, burning sphere of gases) & (sun, is example of, Star) 12. X is Y or Z => (X, is-a, Y) or (X, is-a, Z) freq 1 Force is a push or pull (Force, is-a, push) OR (Force, is-a, pull) 13. X is Y rel1 Z rel2 W => (X, is-a, Y) & (Y, rel1, Z) & (Y, rel2, W) freq 1 Electromagnet is an arrangement of wire wrapped around a core producing a temporary magnet (Electromagnet, is-a, arrangement of wire) & (arrangement of wire, wrapped around, core) & (arrangement of wire, producing, a temporary magnet) 14. X is Y that Z rel1 as it rel2 W => (X, is-a, Y) & (Z, rel1, Y) & (Z, rel2, W) freq 1 Orbit is the path that an object such as a planet makes as it revolves around a second object (Orbit, is-a, path) & (object such as a planet, makes, path) & (object such as a planet, revolves around, a second object) 15. X is Y prep1 Z prep2 W => (X, is-a, Y) & (Y, is prep1, Z) & (Y, is prep2, W) freq 1 Competition is the struggle among living things for the same resources (Competition, is-a, struggle) & (struggle, is among living things) & (struggle, is for, the same resources)