Annotating Agents as Sources

In these annotations, there is one role that an agent can fill, namely that of being the source of a private state or speech event.

In an article, an agent may be referenced any number of times, and may be a source for any number of speech events or private states.

Consider the following sentence.

(1)China said on Tuesday (2)a U.S. State Department report that accused (3)Beijing of suppressing religious freedom was full of lies.

In this example, there are two agents that we are interested in: China and a U.S. State Department report. Also, there are two references to the agent China. These are the (1)China and (3)Beijing spans above.

The annotations for the phrases (1) China, (2) a U.S. State Department report, and (3) Beijing are explained below.

AGENT: China

The sentence begins, "China said." Here China is the agent (according to the writer) that is the source of the speech event indicated by `said'. The span, `China', is annotated as an agent with the following features:

1. id=china
2. nested-source=writer,china

Because this is the first meaningful reference to China, the agent, the `China' span annotation is assigned an identifier (id=china) that will be used to refer to the agent China in any annotation throughout the document. You should NOT add the feature, id, to any another other annotation referencing the agent China, anywhere else in the document.

The nested-source feature of the `China' span annotation indicates that China is the source of the speech event, `said'.

AGENT, target: a U.S. State Department report

The second agent to annotate is the span for the U.S. report.

Notice that the U.S. report is not only a source for the speech event `accused', but it is also the target of the negative emotions (attitude) of China. China thinks the report is "full of lies." So the report is both a source and a target. In the agent annotation, capture only the function of the agent as a source by using the nested-source feature. Also, enter the agent id in the nested-source feature of the DSE `accused'.

Then add a target label on the span `a U.S. State Department report', give it an id, and link it to the writer's negative sentiment attitude. ¹

When indicating that an agent annotation is a nested-source, we maintain the nesting. The report, according to China, according to the writer, is accusing.

Not that in target annotation frames, there is no need to display the nesting of sources. For instance, the fact that the report is full of lies according to China, according to the writer can be derived by `going upstream' from the target 'US State Department report' to the negative-sentiment attitude that it links to, and then on to the DSE `said' that the attitude links to. Inside the annotation frame for `said', we can look at the nested-source feature to determine the nesting.

target: Beijing

This is the second reference to China. China(Beijing) is being accused by the U.S. report of suppressing religious freedom, so it is the target of a negative attitude from the US.
Thus, we have to create a target label for Beijing, give it a unique id, and link it to the negative sentiment attitude that belongs to the DSE `accused' and has the U.S. State Department report as its source.²