This page contains implementation of a method for taxonomy induction that reached the first place in the SemEval 2016 challenge on taxonomy extraction evaluation. The method builds a taxonomy from a domain vocabulary. It extracts hypernyms from substrings and large domain-specific corpora bootstrapped from the input vocabulary. Multiple evaluations based on the SemEval taxonomy extraction datasets of four languages and three domains show state-of-the-art performance of our approach. This page contains implementations of the method including all resources needed to reproduce experiment described in the following paper:

  title={TAXI at SemEval-2016 Task 13: a Taxonomy Induction Method based on Lexico-Syntactic Patterns,  Substrings and Focused Crawling},
  author={Panchenko, Alexander and Faralli, Stefano and  Ruppert, Eugen and Remus, Steffen and  Naets, Hubert and  Fairon, Cedrick and Ponzetto, Simone Paolo and Biemann, Chris},
  booktitle={Proceedings of the 10th International Workshop on Semantic Evaluation},
  address={San Diego, CA, USA},
  organization={Association for Computational Linguistics}

If you would like to refer to the system please use this citation.


TAXI is a taxonomy induction method first presented at the SemEval 2016 challenge on Taxonomy Extraction Evaluation. We consider taxonomy induction as a process that should -- as much as possible -- be driven solely on the basis of raw text processing. While some labeled examples might be utilized to tune the extraction and induction process, we avoid relying on structured lexical resources such as WordNet or BabelNet. We rather envision a situation where a taxonomy shall be induced in a new domain or a new language for which such resources do not exist. Otherwise, there is little need for induction, and in application-based scenarios it is still possible to merge induced and existing taxonomies. In this paper, we demonstrate our methodology by executing hyponymy pattern extraction on general-domain and domain-specific corpora for four languages.


Taxonomy Induction Method

Our approach is characterized by scalability and simplicity, assuming that being able to process larger input data is more important than the complexity of the approach. Our approach to taxonomy induction takes as input a set of domain terms and general-domain text corpora and outputs a taxonomy. It consist of four steps. First, we crawl domain-specific corpora based on terminology of the target domain. These compliment general purpose corpora, like Wikipedia. Second, candidate hypernyms are extracted based on substrings and lexico-syntactic patterns. These candidates are subsequently pruned so that each term has only few most salient hypernyms. The last step performs optimization of the overall taxonomy structure removing cycles and linking disconnected components to the root. Below we present a description of each of these steps. Full description of the method is available in our SemEval paper.

Download Resources

Taxonomy Visualizations

Domain Gold Standard Taxonomy Generated Taxonomy
Science gold_standard_science.png generated_science.png
Food gold_standard_food.jpg generated_food.png
Environment gold_standard_environment.png generated_environment.png

Useful Links


If you have any questions regarding the project write to Alexander Panchenko (email available at or open a Github issue.