Skip to content

Latest commit

 

History

History
120 lines (86 loc) · 4.49 KB

NEW_SYNSETS.md

File metadata and controls

120 lines (86 loc) · 4.49 KB

Guidelines for the addition of new synsets

New synsets should be added with some caution to English WordNet. This document describes the criteria that should be applied before introducing a new synset to the resource.

There are five basic criteria that need to be applied:

  1. Significance
  2. Non-compositionality
  3. Distinction
  4. Well-defined
  5. Linked

All criteria can be ignored in special cases. If you think you have a special case please create an issue.

NB: Many existing synsets do not fit the criteria given below, this is not a reason to add more synsets of poor quality

Significance

A concept in English WordNet should be significant, this means that it should be possible to easily find at least 100 examples of the usage of the word with this meaning. This can be done by using a search interface such as Sketch Engine or other corpus search interface.

NB: At some point we will standardise and integrate this process into the workflow

In the case that a new sense of an existing word is being proposed, then it should be possible to propose collocates that occur with this sense of the word and these can be used to find and distinguish examples.

English WordNet is a dictionary not an encyclopedia. For this reason, it should not contain long lists of people, places, organizations, etc. Proper nouns are generally not expected to be included in the resource and many kinds of common nouns for narrow domains or geographical usage should not be included, examples of this would include elements of different cuisines around the world. As a rule of thumb, if there is a Wikipedia page for this concept it should not be in English WordNet

NB: Deeper integration of Wikipedia/Wikidata is planned to allow these concepts to be referred to from English WordNet

Non-compositionality

One of the goals of English WordNet is to support annotation. If a word or term is already covered by English WordNet it should not be added.

For multiword terms, this means that the meaning of the term should not be derivable from its components, e.g., "French Army" could be tagged with the synsets for "French" and "Army"; in contrast "French Academy" refers not to any French Academy but a specific organization.

For single words, the word should not be derived in a systematic manner, these include:

  • Converting a verb to a noun or adjective by adding '-ing' or '-ed'
  • Converting an adjective to an adverb by adding '-ly'
  • Productive prefixes such as 'non-', 'un-'
  • Systematic polysemy: e.g., using a part to refer to a whole ** Example: "congress" meaning the "members of congress"

Distinction

The concept should be distinct from other concepts in the WordNet. You should think about and check relevant synonyms. This should probably be considered in terms of a substitution check, e.g.,

"happy" and "felicitous" are synonyms, ewn-01052105-s and the examples can be substituted, e.g., "a happy life"/"a felicitous outcome". This does not mean that they can be substituted in every sense, e.g., "happy to help" but not *"felicitous to help".

Well-defined

It should be possible to easily write a definition for this concept that is distinct from other concepts in English WordNet. A good definition consists of a genus and a differentia

  • Genus: The type of the thing, often the hypernym
  • Differentia: Something that makes this word unique

A good definition:

a piece of furniture having a smooth flat top that is usually supported by one or more vertical legs

Bad definition:

a piece of furniture

used for eating

In addition an example should be provided with a link to a website where the example is used as follows:

  <Synset id="ewn-...">
    ...
    <Example dc:source="https://en.wikipedia.org/wiki/Example.com">
    The example domains have one subdomain name defined in the Domain Name System
    </Example>
    ...
  </Synset>

Linked

The synset should be possible to link into the graph

  • Nouns: A hypernym must be identified
  • Verbs: A hypernym or an antonym must be identifier. Verbs should also have at least one subcategorization frame.
  • Adjectives: They should be marked as similar to a non-satellite adjective (in which case they are satellites) or antonyms of a non-satellite adjective or hypernyms of an adjective
  • Adverbs: No clear guidelines but at least one links should be proposed.

The more links that can be provided the better a synset is.