Characteristics of WordNet-Affect 1.1

This version includes a smaller number of synsets but the semantic organization is more well-structured.

a) Affective Hierarchy

The affective label "emotion" is expanded in order to include a subset of new labels, identifying emotional states. These labels, named "affective categories", are hierarchically organized.


b) Valence

The hierarchy was initially obtained from the hyponym subtree of the synset "n#feeling#1", but some modifications were performed in order to classify affective synsets according to emotional valence. In particular, affective categories are partitioned in 4 classes: "positive" (e.g. joy), "negative" (e.g. sadness), "ambiguous" (e.g. surprise), and "neutral" (e.g. apathy).

c) Causative/Stative Attribute

Synsets of part of speech (pos) "adjective", "verb", and "adverb" present an addictional label representing their "causative" or "stative" semantic function. For example, an emotional adjective is "causative" if it refers to some emotion that is caused by the entity represented by the modified noun (e.g. "amusing movie"). On the other hand, an emotional adjective is "stative" if it refers to the emotion owned or felt by the subject denoted by the modified noun (e.g. "cheerful/happy boy").

---------------------------------------------------------------------
			Differences with respect to WordNet-Affect 1.0

- Source files are formatted in XML standard.

- Previous affective labels were renamed and expressed without abbreviations. The mapping between previous and present labels is the following:

	   phy -> physical-state
	   beh -> behaviour
	   sit -> (emotion eliciting) situation
	   tra -> trait
	   sen -> sensation
	   cog -> cognitive-state
	   moo -> mood
	   emo -> emotion
	   eds -> edonic-signal

- We removed the label "core" (referring to manually annotated synsets) and other labels automatically added to synsets (by application of WordNet relations, such as "similar-to"). In fact, all synsets in WordNet-Affect 1.1 were manually reviewed and it is no more useful to trace how they were collected.

- Synsets that are not tagged with the label "emo(tion)" in the previous version are not present in current release. In order to retrieve these synsets, you have to refer to source files of WordNet-Affect 1.0
 
---------------------------------------------------------------------
			File description

a-hierarchy.xml:

Includes the affective hierarchy. Each item has 2 attributes:
	 
	 name = affective category label
	 isa  = category parent in the hierarchy


a-synsets.xml:

Includes synsets associated with the affective hierarchy. Synsets are classified according to their pos. 

Synsets of pos "noun" have the following attributes:

	   id = label identifying current synset
	categ = affective category label

Synsets of other pos ("adjective", "verb", and "adverb") have the following attributes:

	        id = label identifying current synset
	   noun-id = id of the noun synset from which the current one was derived
       causat-stat = causative/stative label

The reason why not-noun synsets are connected to the affective categories via noun synsets is because this relation allows us to study to what extent the causative/stative character of adjectives, verbs and adverbs depends on the morphological variation of nouns. In the next release of WordNet-Affect, it is reasonable to characterize this semantic function of morphology.

---------------------------------------------------------------------
			Plans for the future

The next version of WordNet-Affect will include all synsets that in WordNet-Affect 1.0 are annotated with labels different from "emo" (emotion) and that are not included in the current release. In particular, we want to distinguish labels representing mental states (e.g. cognitive states, attitudes), attributes of mental states (e.g. valence, intensity or level of arousal), and other semantic characteristics (e.c. behaviours, emotion-eliciting situations, emotional responces). Finally, we want to select only one label for each synset, taking into account its hypernyms.