DKPro Core™ Type System Reference

Marks the beginning of a chain.

A link in the coreference chain.

Discourse argument (arg1, arg2)

Attribution annotation (see PTDB for details); not connected to any particular relation as it may belong to two relations thus is covered by DiscourseRelation

Discourse connective

Discourse relation

Discourse relation

RST Tree node

Implicit discourse relation

The DocumentMetaData annotation stores information about a single processed document. There can only be one of these annotations per CAS. The annotation is created by readers and contains information to uniquely identify the document from which a CAS was created. Writer components use this information when determining under which filename a CAS is stored.

There are two principle ways of identifying a document:

• collection id / document id: this simple system identifies a document within a collection. The ID of the collection and the document are each simple strings without any further semantics such as e.g. a hierarchy. For this reason, this identification scheme is not well suited to preserve information about directory structures.
• document base URI / document URI: this system identifies a document using a URI. The base URI is used to derive the relative path of the document with respect to the base location from where it has been read. E.g. if the base URI is file:/texts and the document URI is file:/texts/english/text1.txt, then the relativ path of the document is english/text1.txt. This information is used by writers to recreate the directory structure found under the base location in the target location.

It is possible and indeed common for a writer to initialize both systems of identification. If both systems are present, most writers default to using the URI-based systems. However, most writers also allow forcing the use of the ID-based systems.

In addition to the features given here, there is a language feature inherited from UIMA's DocumentAnnotation. DKPro Core components expect a two letter ISO 639-1 language code there.

A general purpose annotation to store document-wide information in the form of arbitrary key-value string pairs.

Description of an individual tag.

Information about a tagset (controlled vocabulary).

Morphological categories that can be attached to tokens. The features are supposed to match the Universal Dependency v1 features.

The part of speech of a word or a phrase.

Adjective @deprecated Use POS_ADJ instead

Adposition @deprecated Use POS_ADP instead

Adverb @deprecated Use POS_ADV instead

Determiners and articles. @deprecated Use POS_DET instead

at-mention (indicates another user as a recipient of a tweet) @deprecated Use POS_AT instead

Auxiliary verb @deprecated Use POS_AUX instead

Numerals @deprecated Use POS_NUM instead

Conjunction @deprecated Use POS_CONJ instead

Determiner @deprecated Use POS_DET instead

discourse marker, indications of continuation of a message across multiple tweets @deprecated Use POS_DM instead

emoticon @deprecated Use POS_EMO instead

Hashtag (indicates topic/category for tweet) @deprecated Use POS_HASH instead

proper noun + verbal @deprecated Use POS_INT instead

Interjection @deprecated Use POS_INTJ instead

Nouns @deprecated Use POS_NOUN instead

Common noun @deprecated Use POS_NOUN instead

nominal + verbal @deprecated Use POS_NNV instead

Noun @deprecated Use POS_NOUN instead

Proper noun @deprecated Use POS_PROPN instead

proper noun + verbal @deprecated Use POS_NPV instead

Numeral @deprecated Use POS_NUM instead

Catch-all for other categories such as abbreviations or foreign words @deprecated Use POS_X instead

Particle @deprecated Use POS_PART instead

Adjective

Adposition

Adverb

at-mention (indicates another user as a recipient of a tweet)

Auxiliary verb

Conjunction

Determiner

discourse marker, indications of continuation of a message across multiple tweets

emoticon

Hashtag (indicates topic/category for tweet)

proper noun + verbal

Interjection

nominal + verbal

Noun

proper noun + verbal

Numeral

Particle

Pronoun

Proper noun

Punctuation

Subordinating conjunction

Symbol

URL or email address

Verb

Other

Prepositions and postpositions @deprecated Use POS_ADP instead

Pronoun @deprecated Use POS_PRON instead

Pronoun @deprecated Use POS_PRON instead

Proper noun @deprecated Use POS_PROPN instead

Particles @deprecated Use POS_PART instead

Punctuation marks @deprecated Use POS_PUNCT instead

Punctuation @deprecated Use POS_PUNCT instead

Subordinating conjunction @deprecated Use POS_SCONJ instead

Symbol @deprecated Use POS_SYM instead

URL or email address @deprecated Use POS_URL instead

Verbs @deprecated Use POS_VERB instead

Verb @deprecated Use POS_VERB instead

Other @deprecated Use POS_X instead

A document annotation that describes the metadata of a newspaper article.

Represents the phonetic transcription of some textual element (usually a Token). Phonetic transcriptions are e.g. generated by transcription processes like Soundex or Metaphone.

This type represents a decompounding word, i.e.: flowerpot. Each Compound one have at least two Splits.

Document structure element.

This type represents a part of a decompounding word. A Split can be either a CompoundPart or a LinkingMorpheme.

This annotation can be used to indicate an alternate surface form. E.g. some corpora consider a normalized form of the text with resolved contractions as the canonical form and only maintain the original surface form as a secondary information. One example is the Conll-U format.

Token is one of the two types commonly produced by a segmenter (the other being Sentence). A Token usually represents a word, although it may be used to represent multiple tightly connected words (e.g. "New York") or parts of a word (e.g. the possessive "'s"). One may choose to split compound words into multiple tokens, e.g. ("CamelCase" -> "Camel", "Case"; "Zauberstab" -> "Zauber", "stab"). Most processing components operate on Tokens, usually within the limits of the surrounding Sentence. E.g. a part-of-speech tagger analyses each Token in a Sentence and assigns a part-of-speech to each Token.

A alternative token text which should be used instead of the covered text if set on a token.

A CompoundPart represents one fragment from the compounding word. Besides that, it can store other CompoundParts if it can be split again. The way it stores a decompounding word represents a decompounding tree.

Document title, section heading, etc.

This type represents a linking morpheme between two CompoundParts.

Named entities refer e.g. to persons, locations, organizations and so on. They often consist of multiple tokens.

The SemArg annotation is attached to semantic arguments of semantic predicates. Semantic arguments are characterized by their semantic role, e.g. Agent, Experiencer, Topic. The semantic role of an argument is related to its semantic type (for communication verbs, the Agent can be a person or an organization, but typically not food).

The SemArgLink type is used to attach SemPred annotations to their respective SemArg annotations while giving each link a role.

One of the predicates of a sentence (often a main verb, but nouns and adjectives can also be predicates). The SemPred annotation can be attached to predicates in a sentence. Semantic predicates express events or situations and take semantic arguments expressing the participants in these events or situations. All forms of main verbs can be annotated with a SemPred. However, there are also many nouns and adjectives that take arguments and can thus be annotated with a SemanticPredicate, e.g. event nouns, such as "suggestion" (with arguments what and by whom), or relational adjectives, such as "proud" (with arguments who and of what).

The SemanticArgument annotation is attached to semantic arguments of semantic predicates. Semantic arguments are characterized by their semantic role, e.g. Agent, Experiencer, Topic. The semantic role of an argument is related to its semantic type (for communication verbs, the Agent can be a person or an organization, but typically not food). The semantic type of arguments is not yet covered by the SemanticType. @deprecated Use SemArg instead.

The SemanticField is a coarse-grained semantic category that can be attached to nouns, verbs or adjectives. Semantic field information is present e.g. in WordNet as lexicographer file names. Previously, this kind of semantic information has also been called supersenses or semantic types.

One of the predicates of a sentence (often a main verb, but nouns and adjectives can also be predicates). The SemanticPredicate annotation can be attached to predicates in a sentence. Semantic predicates express events or situations and take semantic arguments expressing the participants in these events ore situations. All forms of main verbs can be annotated with a SemanticPredicate. However, there are also many nouns and adjectives that take arguments and can thus be annotated with a SemanticPredicate, e.g. event nouns, such as "suggestion" (with arguments what and by whom), or relational adjectives, such as "proud" (with arguments who and of what). @deprecated use SemPred instead