DKPro WSD

DKPro WSD is a modular, extensible Java framework for word sense disambiguation. It is based on Apache UIMA, an industry standard for text processing.

What is DKPro WSD?

DKPro WSD provides UIMA components which encapsulate corpus readers, linguistic annotators, lexical semantic resources, WSD algorithms, and evaluation and reporting tools. You configure the components, or write new ones, and arrange them into a data processing pipeline.

DKPro WSD is modular and flexible. Components which provide the same functionality can be freely swapped. You can easily run the same algorithm on different data sets, or test several different algorithms on the same data set.

Evaluation and reporting tools

With DKPro WSD, you can…

  • evaluate any number of WSD algorithms simultaneously
  • view or save the raw sense assignments
  • compute common evaluation metrics: coverage, precision, recall, F,,1,, score
  • use binary (“correct” vs. “incorrect”) or real-valued scoring
  • select from various tie-breaking strategies for sense assignments
  • plot precision–recall graphs
  • produce confusion matrices to compare pairs of algorithms
  • filter or break down results by part of speech, etc.
  • show results with and without user-specified backoff algorithms
  • automatically rescore results with a user-specified sense clustering
  • see reports in your web browser, or save them as XML, CSV, etc.

Additional features

  • integrates well with off-the-shelf linguistic UIMA components, such as those from DKPro Core
  • supports parameter sweeping to find optimal system configurations
  • supports related tasks such as entity linking and word sense induction
  • features animated, interactive visualizations of the disambiguation process

Supported resources

DKPro WSD has out-of-the-box support for many different data sets, WSD algorithms, and lexical semantic resources:

  • sense-annotated corpora and data sets
    • Senseval and SemEval tasks (all-words, lexical sample, WSI)
    • AIDA CoNLL-YAGO
    • TAC KBP entity linking tasks
    • MASC
    • SemCor
    • WebCAGe
  • disambiguation algorithms
    • most frequent sense (MFS) baseline
    • (computed) random sense baseline
    • Lesk variants (original, simplified, extended, lexically expanded, …)
    • graph connectivity–based algorithms
    • Personalized PageRank
    • supervised TWSI
    • It Makes Sense (IMS)
  • word sense inventories
    • WordNet 1.7 through 3.1
    • EuroWordNet
    • TWSI
    • Wiktionary
    • Wikipedia
    • GermaNet
    • FrameNet
    • OmegaWiki
    • VerbNet
    • OpenThesaurus
    • DKPro LSR
    • UBY