GF WordNet

GF WordNet is a large interlingual lexicon created by combining resources such as WordNet, PanLex and Wikipedia, plus several translation dictionaries and morphological lexicons. See here for details:

Krasimir Angelov. A Parallel WordNet for English, Swedish and Bulgarian. Proceedings of The 12th Language Resources and Evaluation Conference. 2020. Pages 3008-3015. (PDF)

The lexicon consists of abstract lemmas such as apple_1_N which are mapped for each language to an inflection table. The table contains all possible forms for the most appropriate verbalization of the lemma.

Like in WordNet, the abstract lemmas are organized in synsets which group together lemmas with similar meaning. The synsets, on the other hand, are oganized into a semantic graph copied from WordNet.

The lexicon is compatible with the GF's Resource Grammars Library. Moreover, the grammar can be used to render the examples in the WordNet glosses to different languages. We use the examples to ensure that each abstract lemma is mapped into an appropriate translation for all languages. Since the translations of the examples are generated from a single interlingual abstract syntax tree, it is also possible to compute the alignment between the translations. We show the alignment when the user clicks on a word from the examples.

When there is a corresponding Wikipedia article for a lemma, then we also show the tumbnail image from the corresponding article. Clicking on the image opens the article itself in a new tab.

The lexicon is a work in progress. The status of each verbalization is shown as follows:

• If the verbalization is underlined with a yellow line, then the translation comes from the equivalent synset for the corresponding language. This means that the chosen word has the right meaning but it still might not be the best fitting translation.
• If the verbalization is underlined with a red line then the translation comes from PanLex. This means that the chosen word is a possible translation, but it might be for another sense. When we add a new language, we use the already existing languages as a guide to find the best translation in the target language. Basically we look up all translations in PanLex from each existing language to the target language. After that we pick the translation which fits as many of the existing languages as possible. Often this is enough to also do sense diambiguation but mistakes do happen.
• If there is no special marking then the verbalization has already been validated.

The lexicon contains about 100 000 lemmas, however, only for English all lemmas have a verbalization. The relative lexicon sizes for different languages as well as the current lemma statuses are summarized in the diagram bellow:

The following is a list of all resources in addition to Wikipedia and PanLex that have been used in the creation of the current lexicon: