1. Overview
Reported speech in the form of direct and indirect reported speech is an important indicator of evidentiality in traditional newspaper texts, but also increasingly in the new media that rely heavily on citation and quotation of previous postings, as for instance in blogs or newsgroups. We developed an NLP component in form of a GATE [5] resource that can automatically detect and tag reported speech constructs, in particular the source, reporting verb and content. This is intended as a first module for more sophisticated representation and reasoning with attributed information, such as belief reasoning based on nested belief structures. For one example application, have a look at our Fuzzy Believer [6] system.
For the theoretical background and motivation please have a look at our paper Minding the Source: Automatic Tagging of Reported Speech in Newspaper Articles [7], which appeared at LREC 2008, Marrakech, Morocco. If you use our components or resources, a reference to this paper would be welcome.
2. Prerequisites
As our reported speech tagger comes in form of a number of GATE components, you will obviously need GATE [5] itself. Most of the required pre-processing components are included in the GATE distribution. In particular, you will need (see the pipeline configuration details below): 1. Tokenizer, 2. Sentence Splitter, 3. POS-Tagger, 4. Gazetteer, 5. Morphological Analyzer, 6. Verb Phrase Chunker and 7. MuNPEx Noun Phrase Chunker. Of these, you only need to download the MuNPEx NP Chunker [8] seperately.
3. Documentation
The reported speech tagger is designed to be embedded in more complex pipelines. Here, we describe the minimum requirements for obtaining reported speech annotations. Note that in the following discussion we assume you know how to work with GATE, for tasks like adding a new CREOLE repository or loading new components into a processing pipeline. If you haven't done this before, please read the GATE user's guide [9] first!
3.1. Pipeline Configuration
Our reported speech components are designed to be embedded within a complete GATE analysis pipeline. They rely on annotations added by a number of existing GATE processing resources, in particular tokenization, sentence splitting, part-of-speech tagging, noun phrase chunking, verb grouping, and rudimentary morphological analysis. A complete example for a possible pipeline configuration, with the components needed for complete reported speech tagging, is shown in the image to the right.
Note that our pipeline contains an additional component, the Max NP Transducer, which combines some of the base NPs to complex NPs (PP Attachements). This component is not yet distributed with the MuNPEx chunker. We recommend you use one of the available parsers to improve detection of source NPs.
3.2. Result Annotations
Our components then add annotations for the detected reporting verbs and reported speech constructs. All potential reporting verbs are annotated with their semantic dimensions and their main verb. This annotation is labeled reporting verb group ("RVG") and is generated by the reporting verb marker. All detected reported speech occurrences receive a "Reported Speech" annotation generated by the reported speech finder. Different features of the generated annotation contain detailed information about the reported speech construct, as shown in the following table:
| Name | Value |
|---|---|
| source | source of the reported speech sentence |
| sourceStart | start offset of the source |
| sourceEnd | end offset of the source |
| verb | reporting verb |
| cont1-3 | content, possibly fragmented (up to 3 parts) |
| cont1-3Start | start offset of the content |
| cont1-3End | end offset of the content |
4. Evaluation
Unfortuntely, the manually annotated texts described in our paper are currently not available due to copyright restrictions.
However, the archive below contains a small example text WSJ-Example-Sentences.xml with a number of reported speech sentences from our corpus.
5. Download
Latest version is v1.0 from 28.05.2008. Download:
- the complete gzipped tar archive with all JAPE grammars: ReportedSpeechTagger-1.0.tgz [10]
- our research paper Minding the Source: Automatic Tagging of Reported Speech in Newspaper Articles [7]
- the GNU GPL license [11], under which you can use all the components
If you use our components or resources, please cite our LREC 2008 paper: "Minding the Source: Automatic Tagging of Reported Speech in Newspaper Articles [15]", Proceedings of the Sixth International Language Resources and Evaluation Conference (LREC 2008), Marrakech, Morocco : European Language Resources Association (ELRA), May 28–30, 2008.
6. Feedback
For questions, comments, etc., please use the Forum. [16]
7. Version History
- v1.0 (28.05.2008): initial public release
