This is a living document, describing the conceptual data model used by WikibaseLexeme. It is not a specification of any concrete binding, implementation, mapping, or serialization.
A Lexeme is a lexical element of a language, such as a word, a phrase, or a prefix (see Lexeme on Wikipedia). Lexemes are Entities in the sense of the Wikibase data model. A Lexeme is described using the following information:
- An ID. Lexemes have IDs starting with an "L" followed by a natural number in decimal notation, e.g.
L3746552. These IDs are unique within the repository that manages the Lexeme. The ID can be combined with a repository's concept base URI to form a unique URI for the Lexeme.
- A Lemma for use as a human readable representation of the lexeme, e.g. "run".
- The Language to which the lexeme belongs. This is a reference to a concrete Item, e.g. Q1860 for English.
- The Lexical category to which the lexeme belongs. This is given as a reference to a concrete Item, e.g. Q34698 for adjective.
- A list of Statements to describe properties of the lexeme that are not specific to a Form or Sense (e.g. derived from or grammatical gender or syntactic function).
- A list of Forms, typically one for each relevant combination of grammatical features, such as 2nd person / singular / past tense.
- A list of Senses, describing the different meanings of the lexeme (e.g. "financial institution" and "edge of a body of water" for the English noun bank).
Editorial Note: We should provide some hint regarding how grammatical gender can be modeled using Statements.
In Wikidata they generally use the most general lexical category possible, e.g. affix and then instead describe which type of affix it is using an instance of-statement.
In Wikidata the community decided to have usage examples in one place on the lexeme because then they know where to look for them. They have to demonstrate two properties: form (d:Property:P5830) and sense (d:Property:P6072). They can have multiple examples from different time periods e.g. different centuries and for formality/informality and written/spoken.
The lemma is a human readable representation of the lexeme (see Lemma on Wikipedia). Typically, the canonical form of the lexeme (e.g. the infinitive form of verbs) will be used as the lemma (see also lemon:canonicalForm). Lemmas are not simple strings, but MultilingualTextValues, since the same lemma may have multiple spellings. This is specially important for languages that use multiple scripts such as Serbian and Japanese.
Example: The Lemma for English noun color would include "colour" for British English as well as "color" for American English.
A Lemma cannot be entirely empty, at least one variant has to be provided.
Note: Lemmas are not unique, nor is the combination of Lemma, Language, and Lexical category. Two distinct lexemes with the same lexical category can exist in the same language if they have different data, it may be gender, etymology, morphology (different forms), and so on.
Example: There are two German nouns with the Lemma "See", differing only in gender: "der See" meaning "the lake", and "die See" meaning "the sea". These two meanings cannot be understood as a single Lexeme, since they have different forms according to their gender. An example where there would only be one Lexeme would be the German word "Schild", which for its meaning "Shield" can be either masculine or neutral gendered based on the region in Germany (neutral in the north and parts of central Germany).
The morphology of the lexeme is understood as a set of Forms. Each form defines how a lexeme changes based on a specific syntactic role or mode it may take in a sentence (see also lemon:Form).
Example: The English verb run becomes "running" as a present participle and "runs" in 3rd person singular.
A Form is described using the following information:
- An ID. Forms have IDs starting with the ID of the Lexeme they belong to, followed by a hyphen ("-") and an "F", followed by a natural number in decimal notation: e.g.
L3746552-F7. These IDs are unique within the repository that manages the Lexeme. The ID can be combined with a repository's concept base URI to form a unique URI for the Form.
- A representation, spelling out the Form as a string.
- A list of grammatical features that define for which syntactic role the given form applies. These are given as references to a concrete Items, e.g. Q814722 for participle.
- A list of Statements further describing the Form or its relations to other Forms or Items (e.g. pronunciation audio, rhymes with, used until, used in region)
We may add the notion of a "form type" that determine what information a Form contains. One possible new type could be "nonexistent", which would allow to to represent forms that are known to not exist (like the infinitive of English "may", or the plural of German "Schnee"). Forms of the "nonexistent" kinds would have statements and grammatical features, but no representations.
A form's Representation is its written form, as used in a text (compare lemon:writtenRep). Just like Lemmas, Representations are not simple strings, but MultilingualTextValues, since the same form may have multiple spellings, possibly in multiple scripts.
A Representation cannot be entirely empty, at least one variant has to be provided.
Multiple forms with the same representation are allowed to enable adding usage examples demonstrating each of them. Example in Wikidata
A form's grammatical features specify under which conditions or in which syntactic role that form is used (see lexinfo:morphosyntacticProperty and grammatical category on Wikipedia). Multiple grammatical features can be combined to express under which conditions the language's grammar requires a given form to be used. Grammatical features are represented as references to Items.
Example: The role 1st person present tense plural can be defined by three features, represented by Wikidata Items: Q192613 (present tense), Q21714344 (first person), and Q146786 (plural).
Editorial Note: How do we model "a" vs "an"? What item would we use as a feature to describe this? Do we need free text usage notes after all?
Editorial Note: We should note that gender-specific forms like "baroness" can be treated as Forms, or as separate Lexemes, as need be.
The senses of a lexeme are different meanings which it may represent in a text. The senses are given as natural language definitions or glosses (compare intensional definitions on Wikipedia).
A sense is described using the following information:
- An ID. Senses have IDs starting with the ID of the Lexeme they belong to, followed by a hyphen ("-") and an "S", followed by a natural number in decimal notation: e.g.
L3746552-S4. These IDs are unique within the repository that manages the Lexeme. The ID can be combined with a repository's concept base URI to form a unique URI for the Sense.
- A Gloss, defining the meaning of the Sense using natural language.
- A list of Statements further describing the Sense and its relations to Senses and Items (e.g. item for this sense, synonym, antonym, connotation, register, denotes, evokes). In Wikidata image is also added to provide a culturally adapted image of the sense, e.g. of a letterbox or color that can vary greatly between cultures.
Editorial Note: We should find a good place to address a common source of misunderstandings: Senses can be connected to Wikidata Items via an appropriate Statement they evoke or denote (compare lemon:denotes and lemon:evokes). However, such a connection should not be interpreted as the lexeme actually representing the concept defined by the item (compare lemon:LexicalSense and lemon:LexicalConcept). In particular, if two lexemes have senses that refer to the same concept in this way, this does not imply that the two lexemes are synonyms.
Example: The lexemes for the English adjectives "hot" and "cold" could both have a sense that refers to Q11466 (temperature), even though they are antonyms.
Editorial Note: We should describe how word function can be described for things like "to" or "a", using Statements on the Lexeme. We should also explain that function words should not have senses. Do we need free text usage notes?
We may introduce a field in the Sense for syntactic markers and/or syntactic frames for subcategorization (see also the definition on the UNL wiki). That would allow "ask for", "ask about", "ask to", "ask out", "ask oneself", etc. to be modeled as sense of the same lexeme, each with a different subcategorization. Some verbs also change the meaning depending on whether they are used reflexively (e.g. German "übernehmen" vs "sich übernehmen"). Compare synsem:marker and synsem:syntactic-frame.
A sense's gloss gives a natural definition of the sense (see Gloss on Wikipedia and skos:definition). Glosses cannot be referenced.
Similar to Lemmas, Glosses are not simple strings, but MultilingualTextValues. However, the reason is not providing support for variants, but to allow the gloss to be given in entirely different languages. E.g. it would be quite useful for a German learning French to have a German gloss for a French sense.
A Gloss cannot be entirely empty, at least one language has to be provided. A good gloss provides little or no space for ambiguity about the meaning. Lexemes with multiple senses should have glosses that are easily distinguishable from each other.
Short glosses of only a single or a few words should be avoided as it leaves too much space for interpretation of the meaning.
In Wikidata Glosses are often very similar to carefully crafted descriptions on Q-items. E.g. for apple the Q-items English description fruit of the apple tree is copied as gloss when using tools like MachtSinn to match lexemes and Q-items together and create missing senses.