|
Emotion vocabulary and sentiment analysis in FrameNet| old_uid | 12132 |
|---|
| title | Emotion vocabulary and sentiment analysis in FrameNet |
|---|
| start_date | 2013/02/22 |
|---|
| schedule | 11h-12h30 |
|---|
| online | no |
|---|
| summary | The FrameNet project (Baker, Fillmore & Lowe 1998) has been working for a while now on an analysis of the English vocabulary based on frame semantic theory (Fillmore 1982, 1985; Fillmore & Atkins 1992), documenting and verifying its analyses by creating corpus annotations. In this talk, I will give an overview of FrameNet's method using the example of emotion-related words and then present recent ideas for extending FrameNet's representation for the purposes of sentiment analysis.
In the discussion of emotion vocabulary, I will address several points. First, comparing different releases, we ask how have FrameNet's criteria for distinguishing frames evolved over time? For instance, the organization of emotion frames in the current FrameNet release 1.5 is less exclusively valence-driven than in earlier versions. The overall trajectory is towards an organization where frame divisions are becoming finer and lexical units in a given frame are more semantically similar to each other.
A second key point is that even given the more fine-grained analysis, FrameNet's division of frames and lexical units does not follow any one psychological theory of emotion. However, as will be shown, it is closest in spirit to the work of Ortony, Clore & Collins (1988) whose work is linguistically inspired and whose categories are often compatible with work underway at FrameNet.
In the final part of the talk I suggest ways to extend FrameNet's representation for the purposes of sentiment analysis, producing a resource that allows for deep analysis of not only emotional vocabulary but evaluative and subjective language in general. A key feature of this approach is to piggy-back the extraction of opinion holders and targets onto automatic semantic role labeling systems. |
|---|
| responsibles | Candito |
|---|
| |
|