Abstract. This paper describes a corpus annotation project to study issues in the manual annotation of opinions, emotions, sentiments, speculations, evaluations and other private states in language. The resulting corpus annotation scheme is described, as well as examples of its use. In addition, the manual annotation process and the results of an inter-annotator agreement study on a 10,000-sentence corpus of articles drawn from the world press are presented.

This paper presents this algorithm, gives demonstrations of an implemented system, and describes the results of some preliminary empirical studies, which lend support to the algorithm

Part of understanding fictional narrative text is determhfing for each sentence whether it takes some c, haracter's point of view and, if it does, identifying the character whose point of view is taken. This paper presents part of an algorithm for performing the latter. When faced with a sentence that takes a character's point of view, the reader has to decide whether that character is a previously mentioned character or one mentioned in the sentence. We give particular consideration to sentences about private states, such as seeing and wanting, for which both possibilities exist. Our algorithm is based on regulafities in the ways that texts initiate, continue, and resume a character's point of view, found during extensive examinations of published novels and short stories.

This project continues our interdisciplinary research into computational and cognitive aspects of narrative comprehension. Our ultimate goal is the development of a computational theory of how humans understand narrative texts. The theory will be informed by joint research from the viewpoints of linguistics, cognitive psychology, the study of language acquisition, literary theory, geography, philosophy, and artificial intelligence. The linguists, literary theorists, and geographers in our group are developing theories of narrative language and spatial understanding that are being tested by the cognitive psychologists and language researchers in our group, and a computational model of a reader of narrative text is being developed by the AI researchers, based in part on these theories and results and in part on research on knowledge representation and reasoning. This proposal describes the knowledge-representation and natural-language-processing issues involved in the computational implementation of the theory; discusses a contrast between communicative and narrative uses of language and of the relation of the narrative text to the story world it describes; investigates linguistic, literary, and hermeneutic dimensions of our research; presents a computational investigation of subjective sentences and reference in narrative; studies children’s acquisition of the ability to take third-person perspective in their own storytelling; describes the psychological validation of various linguistic devices; and examines how readers develop an understanding of the geographical space of a story. This report is a longer

The propositional content of a reference is the proposition attributing to the referent the properties that correspond to the nouns and modifiers in the reference (for example, the propositional content of `Mary' is that the referent is named `Mary'). During language comprehension, the hearer or reader must determine the set of beliefs with respect to which the propositional content of a reference is to be understood. In the prototypical case, this set consists of the propositions that she believes that the speaker or writer believes that she and the speaker or writer mutually believe. This paper identifies two contexts in which the propositional content of a specific reference is not understood with respect to this set--- subjective and objective sentences in third-person fictional narrative text---and identifies some implications of this for understanding specific references in these contexts. 1 Introduction Specific references are references to particular entities, for example, `a...

This paper discusses the theoretical background and some of the results of an interdisciplinary, cognitive-science research project on the comprehension of narrative text. The unifying theme of our work has been the notion of a deictic center: a mental model of spatial, temporal, and character information contributed by the reader of the narrative and used by the reader in understanding the narrative. We examine the deictic center in the light of our investigations from the viewpoints of linguistics, cognitive psychology, individual differences (language pathology), literary theory of narrative, and artificial intelligence.

This paper describes work in developing probabilistic classifiers for a discourse segmentation problem that involves segmentation, reference resolution, and belief. Specifically, the problem is to segment a text into blocks such that all subjective sentences in a block are from the point of view of the same agent, and to identify noun phrases that refer to that agent. In our method for developing classifiers (Bruce & Wiebe 1994ab), rather than making assumptions about which variables to use and how they are related, statistical techniques are used to explore these questions empirically. Further, the types of models used in this work can express complex relationships among diverse sets of variables. This work is part of a large project that is in an early stage of development. The contributions of this paper are an illustration of framing a high-level discourse problem in such a way that it is amenable to statistical processing while still retaining its core, and a des...

Recent years have brought a significant growth in the volume of research in sentiment analysis, mostly on highly subjective text types (movie or product reviews). The main difference these texts have with news articles is that their target is clearly defined and unique across the text. Following different annotation efforts and the analysis of the issues encountered, we realised that news opinion mining is different from that of other text types. We identified three subtasks that need to be addressed: definition of the target; separation of the good and bad news content from the good and bad sentiment expressed on the target; and analysis of clearly marked opinion that is expressed explicitly, not needing interpretation or the use of world knowledge. Furthermore, we distinguish three different possible views on newspaper articles – author, reader and text, which have to be addressed differently at the time of analysing sentiment. Given these definitions, we present work on mining opinions about entities in English language news, in which (a) we test the relative suitability of various sentiment dictionaries and (b) we attempt to separate positive or negative opinion from good or bad news. In the experiments described here, we tested whether or not subject domain-defining vocabulary should be ignored. Results showed that this idea is more appropriate in the context of news opinion mining and that the approaches taking this into consideration produce a better performance. 1.

Most sentiment analysis work has been carried out on highly subjective text types where the target is clearly defined and unique across the text (movie or product reviews). However, when applying sentiment analysis to the news domain, it is necessary to clearly define the scope of the task, in a more specific manner than it has been done in the field so far. The main tasks we identified for news opinion mining are: definition of the target; separation of the good and bad news content from the good and bad sentiment expressed on the target; and analysis of clearly marked opinion that is expressed explicitly, not needing interpretation or the use of world knowledge. It is furthermore important to distinguish three different possible views on newspaper articles – author, reader and text. These have to be addressed differently at the time of analysing sentiment, especially the case of author intention and reader interpretation, where specific profiles must be defined if the proper sentiment is to be extracted.

The exponential growth of the subjective information in the framework of the Web 2.0 has led to the need to create Natural Language Processing tools able to analyse and process such data for multiple practical applications. They require training on specifically annotated corpora, whose level of detail must be fine enough to capture the phenomena involved. This paper presents EmotiBlog – a finegrained annotation scheme for subjectivity. We show the manner in which it is built and demonstrate the benefits it brings to the systems using it for training, through the experiments we carried out on opinion mining and emotion detection. We employ corpora of different textual genres –a set of annotated reported speech extracted from news articles, the set of news titles annotated with polarity and emotion from the SemEval 2007 (Task 14) and ISEAR, a corpus of real-life selfexpressed emotion. We also show how the model built from the EmotiBlog annotations can be enhanced with external resources. The results demonstrate that EmotiBlog, through its structure and annotation paradigm, offers high quality training data for systems dealing both with opinion mining, as well as emotion detection.