Event Attribute Annotation
Annotation scheme
Each event is assigned three attributes, corresponding to different aspects of its interpretation. Where possible, attribute-specific cue phrases are annotated and linked to the event. These are words/phrases in the same sentence as the event, which are used by the annotator to determine the chosen (non-default) value for a given attribute. Table 1 provides definitions of each of these three attributes, enumerates the possible attribute values and provides some annotated examples, including cue phrases.
Note that in the brat visualisation of the corpus, events that have one or more attributes assigned to them are denoted by a red hash (#).
Attrbute name | Description | Possible Values |
---|---|---|
Negated | Denotes whether or not there is explicit evidence in the sentence that the event should be negated |
True: Explicit negation evidence is present False: There is no evidence that the event should be negated (default value) |
![]() ![]() ![]() |
||
Speculated | Denotes whether there is some degree of uncertainty or speculation as to whether the event will actually take place |
True: There is explicit evidence of one of the following:
|
![]() ![]() ![]() ![]() ![]() |
||
Manner | Denotes whether the manner of the event (i.e., the rate, intensity, strength or level of significance of the event) is higher or lower than would be expected by default |
Low: There is explicit indication that the manner of the event is lower than would be expected by default, e.g. it happens with low intensity, happens rarely or is not considered to be significant High: There is explicit indication that the manner of the event is lower than would be expected by default, e.g. it happens with high intensity, it happens very frequently, or is considered to be very significant. Neutral: There is no explicit indication that the manner of the event is high or lower than would be expected by default (default value) |
![]() ![]() ![]() |
Event attribute statistics
The total number of events for which non-default attribute values are assigned
Attribute | Total number of events with non-default attributes | Number of unique cues |
---|---|---|
Negated | 61 | 19 |
Speculated | 448 | 117 |
Manner | 95 | 47 |
Agreement
The annotations were undertaken by annotators with domain expertise. The quality and consistency of the annotations were verified through the calculation of inter-annotator agreement (IAA) on one quarter of the complete corpus (i.e., 150 abstracts). We calculated IAA in terms of F-Score, as shown in Table 3.
Relation Type | Agreement Rate |
---|---|
Negated | 75.6 |
Speculated | 76.5 |
Manner | 58.9 |
TOTAL | 70.9 |