*****************************************************************************
* Wiltrud Kessler, 2012.
* Institute for Natural Language Processing
* University of Stuttgart       
* http://www.ims.uni-stuttgart.de/~kesslewd
*****************************************************************************
* This work is licensed under a Creative Commons Attribution 3.0 Unported License.
* CC-BY 3.0:
* You are free:
*     to Share — to copy, distribute and transmit the work
*     to Remix — to adapt the work
*     to make commercial use of the work
* Under the following conditions:
*     Attribution — You must attribute the work in the manner specified by the
*  			author or licensor (but not in any way that suggests that they 
* 				endorse you or your use of the work). 
*****************************************************************************
* 
* Please cite this paper if you use these constructions:
* Wiltrud Kessler and Hinrich Schütze (2012)
* Classification of Inconsistent Sentiment Words Using Syntactic Constructions.
* In Proceedings of the 24th International Conference on Computational Linguistics
* (COLING 2012)
* Mumbai, India, 10.-14. December 2012.
* 
*****************************************************************************
* 
* === Format of Constructions ===
* 
* Syntactic constructions describe the syntactic context of a sentiment word.
* We define a syntactic construction as any path 
* that starts at a sentiment word,
* ends at another word in the sentence,
* and contains the POS categories of all nodes that are traversed on the path.
* 
* We parsed all training examples with the Bohnet dependency parser.
* The parts of speech (POS) produced by the parser are generalized to the categories
* N (noun), V (verb), ADJ (adjective), ADV (adverb), 
* PR (preposition), DT (determiner), and * (everything else).
* We evaluated syntactic constructions extracted from the context of a sentiment word
* up to a parse tree distance 3 (defined in number of nodes on the path).
* 
* As an example, consider the syntactic construction 'N<V<V>additionally_ADV'
* in the sentence "additionally I have had no problems with battery life"
* (for the parse tree please refer to the paper).
* The sentiment word "problems" is represented in the syntactic construction by
* POS category 'N', but the word is not included, as we are interested in constructions
* that are independent of specific sentiment words.
* The path contains the direction in the parse tree (up '<' or down '>'),
* the nodes that are traversed on the way 
* -- represented by POS category 'V' for "had" and "have") --
* and lemma/POS of the final word on the path ("additionally", 'ADV').
* 
* === Annotation ===
* 
* We define a  polarity reversing construction (PRC) as a syntactic construction
* that reverses the polarity of the sentiment word in its
* scope. The sentiment word is the first node of
* the path represented by the construction.
* We extracted candidate constructions for PRCs automatically from the
* context of inconsistent words as described in the paper.
* The results of experiments over a period of ~6 months were given to a
* human annotator to annotate.
* The annotator was shown the construction to be annotated and a list of usage examples 
* from the corpus. As the examples were automatically parsed and the sentiment
* words identified and classified as reversed, there were errors in this list.
* The annotator was asked to annotate a candidate as "REVERSER" if in the
* vast majority of usage cases it is the responsible construction for 
* reversing the polarity of the presented sentiment word.
* This list contains all syntactic constructions that have been annotated
* as "REVERSER" by a human annotator.
* 
*****************************************************************************

ADV<not_ADV	REVERSER
ADV<V<V>not_ADV	REVERSER
ADJ<V>not_ADV	REVERSER
ADV<V<would_*	REVERSER
ADV<*>not_ADV	REVERSER
ADV<ADJ>not_ADV	REVERSER
N>no_DT	REVERSER
N>low_ADJ	REVERSER
ADJ>not_ADV	REVERSER
ADJ<N>not_ADV	REVERSER
ADJ<N>no_DT	REVERSER
ADV<V>not_ADV	REVERSER
ADJ<N<without_PR	REVERSER
ADV<ADJ<V>not_ADV	REVERSER
ADJ<V<V>not_ADV	REVERSER
ADJ>only_ADV	REVERSER
ADJ<not_ADV	REVERSER
ADJ>less_ADV	REVERSER
ADJ<V<would_*	REVERSER
V<V>not_ADV	REVERSER
ADJ<V<could_*	REVERSER
ADJ<V<V<would_*	REVERSER
V<would_*	REVERSER
ADJ<N<V>not_ADV	REVERSER
ADJ<V<if_PR	REVERSER
ADJ<N<V<would_*	REVERSER
ADJ<V<*>not_ADV	REVERSER
N<PR<ADJ>not_ADV	REVERSER
N<without_PR	REVERSER
N>not_ADV	REVERSER
N<V<would_*	REVERSER
ADJ>no_DT	REVERSER
ADJ<N<PR<lack_N	REVERSER
N<V>never_ADV	REVERSER
V>not_ADV	REVERSER
ADJ<N<PR<less_ADJ	REVERSER
N<PR<lack_N	REVERSER
V<*<hard_ADJ	REVERSER
V<without_PR	REVERSER
V>never_ADV	REVERSER
V<*<leave_V	REVERSER
N<V>bad_ADJ	REVERSER
V<V<*<leave_V	REVERSER
N<PR<problem_N	REVERSER
N<PR<N>bad_ADJ	REVERSER
N<slow_ADJ	REVERSER
V<*<*<leave_V	REVERSER
N<N<V>bad_ADJ	REVERSER
N<break_V	REVERSER
N<slow_V	REVERSER
ADV<V<V>poor_ADJ	REVERSER
N<*<N>bad_ADJ	REVERSER
N<problem_N	REVERSER
ADV<PR>N>poor_ADJ	REVERSER
N<ADV<problem_N	REVERSER
ADV<N>poor_ADJ	REVERSER
V<PR<delay_V	REVERSER
ADV<PR<N>poor_ADJ	REVERSER
N<PR<slow_ADJ	REVERSER
N<N<problem_N	REVERSER
ADV<N<V>poor_ADJ	REVERSER
N<PR<V<problem_N	REVERSER
V<PR<delay_N	REVERSER
ADV<PR>V>poor_ADJ	REVERSER
N<N>N>slow_ADJ	REVERSER
N<V>bad_N	REVERSER
V<delay_N	REVERSER
N<*<break_V	REVERSER
ADV<too_ADV	REVERSER
V<*>not_ADV	REVERSER