Harnessing Disagreement for Event Semantics
1 like1,245 views
The document discusses the complexities of event semantics, highlighting the differences in human interpretations of what constitutes an event. It emphasizes the challenges posed by disagreement among annotators regarding the classification, granularity, and interpretation of events, which complicates the task of teaching machines to understand these nuances. The authors advocate for a framework that accepts and analyzes disagreement to improve understanding and processing of events in natural language processing.
Harnessing Disagreement for Event Semantics
- 1. Harnessing Disagreement for Event Semantics Lora Aroyo Chris Welty
- 2. Objects vs. Events events perdure = their parts exist at different time points objects endure = they have all their parts at all points in time objects are wholly present at any point in time, events unfold over time Flickr: vanilllaph
- 3. Extraction i n NLP Event • Find Events mentioned in text • Type them (communication, bombing, …) • Find the role fillers (location, date, participants, …) • State of the art is extremely low: .10 F
- 4. Measuring • Define the task for human annotators • Annotate by multiple people • Measure agreement • While k<.6 refine definition and repeat ... for events this process is long, disagreement is high, agreement becomes forced
- 5. Position eir semantics are par t of th t about events en man di sagreem Hu Flickr: elkabong
- 6. Events are Vague Humans have no clear notion of what events are
- 7. event is a significant "happening" or gathering of people. I would define a "happening" as an event if the group of people gathered were united in one common goal. We Asked the Crowd What an EVENT is Flickr: massimo vitali
- 8. Event is a happening, which can be scheduled or unscheduled. An earthquake or fire happens (unscheduled). A wedding or birthday party (scheduled). It is an occasion that is unusual and tends to be memorable. We Asked the Crowd What an EVENT is
- 9. An event would be any occurrence where physical action has taken place. It may be a single, momentary instance (I sneezed), or it may span a period of time (the festival ran for four hours). An event may also be made up of a number of smaller events, such as a day at school is an event, but each individual class is also an event itself. Basically an event must have a physical action over any delimited time span. We Asked the Crowd What an EVENT is
- 10. Event can refer to many things such as: An observable occurrence, phenomenon or an extraordinary occurrence. an event is an incident that's very important or monumental A planned public or social get together or occasion. An event is something occurring at a specific time and/or date to celebrate or recognize a particular occurrence. a location where something like a function is held. you could tell if something is an event if there people gathering for a purpose. We Asked the Crowd What an EVENT is
- 11. What do Experts think an EVENT is? an event is the exemplification of a property by a substance at a given time Jaegwon Kim, 1966 events are changes that physical objects undergo Lawrence Lombard, 1981 events are properties of spatiotemporal regions , David Lewis, 1986 under30ceo.com
- 12. What do Experts think an EVENT is? an event is the exemplification of a property by a substance at a given time Jaegwon Kim, 1966 events are changes that physical objects undergo Lawrence Lombard, 1981 nothing everything events are properties of spatiotemporal regions , David Lewis, 1986 under30ceo.com
- 13. Why is event semantics hard?
- 14. the World is Open 1. events have multiple dimensions 2. each dimension has levels of granularity 3. people have different views on both all this leads to very complex semantics
- 15. and our goal is ... 1. not to enforce agreement 2. to capture different view points 3. to teach machines to reason in the disagreement space Flickr: elkabong
- 16. Artificially Hypothesis restricting humans d Machines oes not h will learn elp mach from dive ines to le rsity arn. Flickr: elkabong
- 17. What do People Disagree on? are sub-events always mere parts? are mentions meaningful for events? are events coreferential across documents? (e.g. perspectives, observations)
- 18. the bombing targeted a housing development in Baghdad, killing 3 and injuring 13 indistinguishable by people, confusable: is bombing part of killing, or killing part of bombing? What about targeting? merelogically extensional (i.e arbitrary): container bursting into fragments as a result of explosion some events don t exist: an action by military forces prevented the bombing.
- 19. Disagreement Framework • ontology: disagreements on the basic status of events themselves as referents of linguistic utterances, e.g. are people events or do events exist at all. • granularity: disagreements that result from issues of granularity, e.g. the location being a country, region, or city, the time being a day, week, month, etc. • interpretation: disagreements that result from (non- granular) ambiguity, differences in perspective, or error in interpreting an expression, e.g. classifying a person as a terrorist/hero, October Revolution took place in September.
- 20. Granularity Disagreement • spatial, temporal, participants • compositional, classificational
- 22. Event Participants Disagreement Israeli Prime minister 50% Government 10% Benjamin Netanyahu Israeli Cabinet 15% his Cabinet 15% 35% Benjamin {TOLD} Netanyahu Benjamin Israeli Prime Netanyahu s 5% 15% minister Cabinet Cabinet 45%
- 23. Temporal Disagreement Prime minister 50% Benjamin Sunday 50% Netanyahu March 1, 1998 25% 35% Benjamin {TOLD} March 1998 15% Netanyahu Spring 1998 5% Israeli Prime 15% minister
- 24. Spatial Disagreement Southern 35% 30% Israel Lebanon {WILLING TO WITHDRAW} 65% Israel's Northern Lebanon 45% Frontier Middle East 10%
- 25. Approach Principles 1. tolerate, capture & exploit disagreement 2. understand the range of disagreements by creating a space of possibilities with frequencies & similarities 3. score machine output based on where it falls in this space 4. adaptable to new annotation tasks Flickr: auroille
- 26. Position eir semantics are par t of th t about events en man di sagreem Hu Flickr: elkabong
- 27. Artificially Conclusion restricting humans d Machines oes not h will learn elp mach from dive ines to le rsity arn. Flickr: elkabong