In the latest edition of the trade publication E-Data & Research, a nice article (in Dutch) about our research on knowledge graphs for maritime history is published. Thanks to Mathilde Jansen and of course my collaborators Stijn Schouten and Marieke van Erp! The image below shows the print article, the article can be found online here.
[This post is the text of a 1-minute pitch at the IWDS symposium for our poster “A Polyvocal and Contextualised Semantic Web” which was published as the paper”Erp, Marieke van, and Victor de Boer. “A Polyvocal and Contextualised Semantic Web.” European Semantic Web Conference. Springer, Cham, 2021.”]
Knowledge graphs are a popular way of representing and sharing data, information and knowledge in many domains on the Semantic Web. These knowledge graphs however often represent singular -biased- views on the word, this can lead to unwanted bias in AI using this data. We therefore identify a need a more polyvocal Semantic Web.
So. How do we get there?
- We need perspective-aware methods for identifying existing polyvocality in datasets and for acquiring it from text or users.
- We need datamodels and patterns to represent polyvocal data information and knowledge.
- We need visualisations and tools to make the polyvocal knowledge accessible and usable for a wide variety of users, including domain experts or laypersons with varying backgrounds.
In the Cultural AI Lab, we investigate these challenges in several interrelated research projects, but we cannot do it, and should not do it alone and are looking for more voices to join us!
Place names (toponyms) are very ambiguous and may change over time. This makes it hard to link mentions of places to their corresponding modern entity and coordinates, especially in a historical context. We focus on historical Toponym Disambiguation approach of entity linking based on identified context toponyms.
The thesis specifically looks at the American Gazetteer. These texts contain fundamental information about major places in its vicinity. By identifying and exploiting these tags, we aim to estimate the most likely position for the historical entry and accordingly link it to its corresponding contemporary counterpart.
Therefore, in this case study, Bram Schmidt examined the toponym recognition performance of state-of-the-art Named Entity Recognition (NER) tools spaCy and Stanza concerning historical texts and we tested two new heuristics to facilitate efficient entity linking to the geographical database of GeoNames.
We tested our method against a subset of manually annotated records of the gazetteer. Results show that both NER tools do function insufficiently in their task to automatically identify relevant toponyms out of the free text of a historical lemma. However, exploiting correctly identified context toponyms by calculating the minimal distance among them proves to be successful and combining the approaches into one algorithm shows improved recall score.
[This blog post is co-written with Marieke van Erp and Rinke Hoekstra and is cross-posted from the Clariah website]
Linked Data, RDF and Semantic Web are popular buzzwords in tech-land and within CLARIAH. But they may not be familiar to everyone within CLARIAH. On 12 september, CLARIAH therefore organized a workshop at the Vrije Universiteit Amsterdam to discuss the use of Linked Data as technology for connecting data across the different CLARIAH work packages (WP3 linguistics, WP4 structured data and WP5 multimedia).
The goal of the workshop was twofold. First of all, to give an overview from the ‘tech’ side of these concepts and show how they are currently employed in the different work packages. At the same time we wanted to hear from Arts and Humanities researchers how these technologies would best suit their research and how CLARIAH can support them in familiarising themselves with Semantic Web tools and data.
Monday afternoon, at 13:00 sharp, around 40 people showed up for the workshop at the Boelelaan in Amsterdam. The workshop included plenary presentations that laid the groundwork for discussions in smaller groups centred around the different types of data from the different WPs (raw collective notes can be found on this piratepad).
Rinke Hoekstra presented an Introduction Linked Data: What is it, how does it compare to other technologies and what is its potential for CLARIAH. [Slides]
In the discussion that followed, some concerns about the potential for Linked Data to deal with data provenance and data quality were discussed.
After this, three humanities researchers from each of the work packages discussed experiences, opportunities, and challenges around Linked Data. Our “Linked Data Champions” of this day were:
- WP3: Piek Vossen (Vrije Universiteit Amsterdam) [Slides]
- WP4: Richard Zijdeman (International Institute of Social History)
- WP5: Kaspar Beelen and Liliana Melgar (University of Amsterdam) [Slides]
Marieke van Erp, Rinke Hoekstra and Victor de Boer then discussed how Linked Data is currently being produced in the different work packages and showed an example of how these could be integrated (see image). [Slides]. If you want to try these out yourself, here are some example SPARQL queries to play with.
Break out sessions
Finally, in the break out sessions, the implications and challenges for the individual work packages were further discussed.
- For WP3, the discussion focused on formats. There are manynatural language annotation formats used, some with a long history, and these formats are often very closely connected to text analysis software. One of the reasons it may not be useful to WP3 to convert all tools and data to RDF is that performance cannot be guaranteed, and in some cases has already been proven to not be preserved when doing certain text analysis tasks in RDF. However, converting certain annotations, i.e. end results of processing to RDF could be useful here. We further talked about different types of use cases for WP3 that include LOD.
- The WP4 break-out session consisted of about a dozen researchers, representing all working packages. The focus of the talk was on the expectations of the tools and data that were demonstrated throughout the day. Various persons were interested to apply QBer, the tool that allows one to turn csv files into Linked Data. The really exciting bit about this, is that the interest was shared by persons outside WP4, thus from persons usually working with text or audio-video sources. This does not just signal the interest in interdisciplinary research, but also the interest for research based on various data types. A second issue discussed was the need for vocabularies ((hierarchical) lists of standard terms). For various research fields such vocabularies do not yet exist. While some vocabularies can be derived relatively easily from existing standards that experts use, it will prove more difficult for a large range of variables. The final issue discussed was the quality of datasets. Should tools be able to handle ‘messy’ data? The audience agreed that data cleaning is the responsibility of the researcher, but that tools should be accompanied by guidelines on the expected format of the datafile.
- In the WP5 discussion, issues around data privacy and copyrights were discussed as well as how memory institutions and individual researchers can be persuaded to make their data available as LOD (see image).
The day ended with some final considerations and some well-deserved drinks.