Wetzoek uses artificial intelligence to make Dutch law more accessible. We make it easier for people without a legal background to find the context for their legal questions. We also give lawyers and other people working with Dutch law a tool to find relevant legislation and case law.
The field of "Natural Language Processing" applies artificial intelligence to natural languages. It contains a wide range of tools that can be applied to the legal space. Examples include the automatic drafting of contracts, the analysis of contracts to identify the most important clauses, or the use of language analysis to arrive at a (preliminary) verdict.
Search engines also make increasing use of artificial intelligence. Traditional search engines retrieved keywords, by means of so-called "symbolic search" . This means that they simply browse through documents and make a note of when one or more keywords appear. This approach has often left little room for nuance: searching for the text "summary dismissal", would mean that the text "he was immediately fired" was not picked up by the search engine.
Most modern search engines are usually equipped to recognise such variations on the desired search terms. This often requires explaining a few rules to the search engines: for example, that "dismissal" is a synonym for "fired". You can imagine that this is time-consuming work that leaves a lot of room for error. A small omission on the developer's part means the search engine no longer functions as effectively.
The latest development at the overlap between search engines and artificial intelligence is neural search. This involves training models on huge amounts of text to recognise similarities, parallels and patterns. A search engine based on neural search uses these models to place search terms or questions in their broader context, in order to come up with a better answer. Wetzoek uses the neural search techniques of open-source companies such as DeepSet and Jina.
Another type of artificial intelligence we use is "multi-label classification" . This involves training a model on labelled texts, so that it can use pattern recognition to give a label to new input text. For example: you feed the machine summaries of "Soldaat van Oranje", "Spetters", "Turks Fruit", and a hundred other films. You have already categorised these summaries: "War film", "Drama". If all goes well, the model will then categorise a summary of the film "Pastorale 1943" as "War film". When using this technique, it is not a requirement that Rutger Hauer has a role in the film.
Wetzoek uses multi-label classification to determine the area of law applicable to input texts. For an initial list of areas of law we use the current definitions used by the official Dutch record of verdicts and cases. With the help of graph theory we can provide even more granular categorisation. You can read more about how we do this in a Medium article: Using Graphs and Neo4j to build a search tool for Dutch law.
Wetzoek would not have been possible with out the excellent Open Data provided by the Dutch government. Both the case law found on rechtspraak.nl and the legal code found on wetten.nl are readily available and form the cornerstone of Wetzoek.