Thatcher’s Ethos in Hansard corpus
The Thatcher’s Ethos in Hansard corpus was constructed by taking a random subsample of Hansard according to the following rubric: select the first two House of Commons debates over 700 words in length from the day closest to the date(s) at the midpoint(s) of the largest uninterrupted date range(s) (initially the midpoint in the range 4th May 1979 and 22nd November 1990 – viz., 11th February 1985; then at the midpoints between 4th May 1979 and 11th February 1985, and between 11th February 1985 and 22nd November 1990, etc.). This avoids bias for annotators and yielded 60 transcripts, the data in each of which was then cleaned such that any titles and section markers were removed to leave only the speakers, organisations or other entities and the statements they made. The transcripts were then split evenly to give a training set and a testing set. The training set formed the training data for the sentiment polarity classifier and was used as the basis for developing domain specific rules for recognising ethotic sentiment expressions.
ESE+/- Annotation scheme
Ethos support. Ethos support should be identified when: (a) the statement makes explicit mentions of a person, organisation or other entity (excluding groups and assemblages) except when this is reported speech; and (b) it takes the form of supporting a person’s credibility or looking to put them in a positive frame through character supports or supports of work; and (c) a support to a person’s own ethos should not be analysed as this is deemed to be a fallacy. Ethos attack. Ethos attack should be identified when: (a) the statement makes explicit mentions of a person, organisation or other entity (excluding groups and assemblages) except when this is reported speech; and (b) it takes the form of attacking a person’s credibility or looking to put them into a negative frame; or (c) it may take the form of trying to unbalance authority on a subject giving the attacker more of a right to talk about the subject.
A 12% of the data was annotated by two annotators for the purpose of evaluation. Overall this gave Cohen’s kappa=0.42, a percentage agreement of 57\% and weighted $\kappa=0.63$. The kappa score is considered fair but the weighted kappa score shows room for improvement in the guidelines. To this extent a third annotator annotated a smaller 8% subset of the data to provide a pairwise agreement score. Comparing all three annotators gave Fleiss kappa = 0.51. A pairwise comparison of annotators one and three gave kappa = 0.63 and weighted kappa = 0.79. Comparing annotators two and three gave kappa = 0.51 and weighted kappa = 0.70.
WVG Annotation scheme
Building upon the Aristotelian distinction between ethos types we proposed the annotation of ethos types using three main tags (Practical Wisdom, Moral Virtue and Goodwill) split into support and attack (Argument and Conflict) according to the following guidelines: Practical Wisdom: Argument From Practical Wisdom should be annotated when: (a) an entity is said to have sufficient knowledge for the purpose at hand; or (b) an entity can draw conclusions from this knowledge; or (c) an entity has practical experience; (d) an entity can draw conclusions from this experience. While Conflict From Practical Wisdom should be annotated when the opposite is true. Moral Virtue: Argument From Moral Virtue should be annotated when: (a) a statement refers to the character trait of an entity, when the entity shows positive morality, calmness, justness, selflessness, gracefulness, nobility, positive contributions, liberality, magnanimity or magnificence; or (b) when an entity provides the correct information. While Conflict From Moral Virtue should be annotated when the opposite is true. Goodwill: Argument From Goodwill should be annotated when: (a) a statement refers to an entity’s ability to show goodwill to others; or (b) an entity gives sound advice when it is know, ensuring the entity does not deceive while being inclusive; or (c) an entity aligns with an audience’s values, displaying self sacrifice. While Conflict From Goodwill should be annotated when the opposite is true.