TL;DR
This paper introduces the Swedish PoliGraph, a semantic graph designed to enhance argument mining in Swedish parliamentary debates. By integrating parliamentary data with named entity recognition, it aims to improve the accuracy of identifying who arguments are directed towards, paving the way for deeper discourse analysis.
What Problem Does This Paper Address?
Argument mining, the task of automatically identifying arguments within text, is a growing field. Parliamentary debates are a rich source of data for this task due to their structured language and accompanying metadata. However, accurately understanding the relationships between arguments and identifying their targets within these complex discussions remains a challenge. The authors aim to address this by creating a structured representation of parliamentary data that facilitates more precise argument analysis and named entity resolution.
Methods
The core of the Swedish PoliGraph is a deductive database implemented in SWI-Prolog, coupled with the Pengines framework for web access. This approach allows for rapid prototyping and flexible querying, leveraging Prolog's declarative nature. The graph integrates various data sources, including:
- Parliamentary Speeches: Transcripts of debates from 1971 to the present, with detailed metadata available from 1993 onwards.
- Member of Parliament (MP) Data: Biographical information, party affiliation, and a history of their roles and activities within parliament.
- Wikidata Integration: Mappings to MPs' Wikidata IDs are created to leverage external knowledge and potentially resolve ambiguities.
The system uses a set of specialized Prolog rules to disambiguate names and titles, enabling queries about MP roles, speech content, and temporal relationships within debates.
Datasets
The primary dataset used is the Swedish parliamentary data from the Riksdagens öppna data (RÖD) initiative. This includes:
- Documents: A large category containing various document types.
- MPs: Metadata about Members of Parliament.
- Voting Results: Information on voting sessions.
- Speeches: Transcripts of addresses and replies, with extensive metadata.
- Calendar: Information on parliamentary meetings.
The authors specifically focus on the "Speeches" category due to its consistency and rich metadata. While the paper doesn't introduce a new benchmark dataset, it leverages and structures existing parliamentary data for a novel analytical purpose.
Key Insights and Ideas
- Semantic Graph for Discourse: The paper highlights the power of representing complex, interconnected data like parliamentary debates as a semantic graph. This structure moves beyond simple text analysis to capture relationships between entities (MPs), events (speeches, debates), and their temporal and topical context.
- Bridging NER and Argument Mining: The Swedish PoliGraph is explicitly designed to improve Named Entity Recognition (NER) as a precursor to more accurate argument mining. By resolving MP identities and their roles, the system can better determine the intended audience of arguments.
- Prolog's Suitability for Complex Queries: The choice of SWI-Prolog demonstrates its effectiveness for building knowledge bases and handling intricate logical relationships, especially when compared to traditional relational databases for this type of task.
- Leveraging Open Data: The project capitalizes on the availability of Swedish parliamentary open data, showcasing a practical application of such initiatives for advanced linguistic and political science research.
Key Results
The paper focuses on the creation and architecture of the Swedish PoliGraph rather than presenting quantitative results from argument mining experiments. The key "result" is the successful development of a structured, queryable semantic graph that integrates diverse parliamentary data. This graph is made available for use and download, enabling researchers to perform complex queries related to MPs, their roles, and their speeches.
Limitations and Future Work
The authors acknowledge that the web interface is simplified and primarily for demonstration, limiting its functionality compared to direct Prolog querying. Future work includes:
- NER Enhancement: Further developing NER capabilities using the PoliGraph.
- Argumentation Mining: Applying the graph to improve argument mining techniques.
- Metadata Augmentation: Potentially adding more metadata to the graph to enrich argumentation mining features.
- Broader Applicability: The graph is intended to be useful for various projects involving Swedish MPs and debates, beyond the authors' immediate research scope.
Summary updated