Relations in GermaNet
Description Navigation
This page introduces the set of relations used in GermaNet. Additional information about the use of relations for the modelling of a particular word class can be found on the page for the word class.
Overview
Concepts are the primary lexicographic unit in GermaNet, and are represented as sets of synonyms (synsets). Each synset denotes a distinct concept and consists of one or more lexical units. A lexical unit represents a specific sense of a word, which is stored in its base form.
There are two types of relations, conceptual and lexical. Conceptual relations denote relations that hold for the full concept, not only for one of its synonyms. Lexical relations hold between individual lexical units (synonyms). Antonymy, for example, is a frequently occurring lexical relation.
Note that relations in GermaNet differ from those in WordNet® in two important aspects:
- GermaNet uses the hyperonymy / hyponymy relations for adjectives, rather than the 'similar-to' relation used in WordNet®.
- Cross-classification (multiple inheritance) is explicitly allowed in GermaNet and lexicographers are encouraged to make use of it whenever appropriate.
Relation | Definition | Example | Classes |
---|---|---|---|
hypernymy (has_hypernym) | from a more specific to a superordinate / more general term | dog —> pet | N A V |
hyponymy (has_hyponym) | from a more general to a more specific / subordinate term | dog —> labrador | N A V |
component-object meronymy (has_component_meronym) | from objects to their components | car —> body | N |
component-object holonymy (has_component_holonym) | from components to their objects | wall —> room | N |
substance-object meronymy (has_substance_meronym) | from a (whole) object to its substance/material | barrel —> wood | N |
substance-object holonymy (has_substance_holonym) | from a substance/material to a (whole) object | oxygen —> water | N |
member-collection meronymy (has_member_meronym) | from groups to their members | nobility —> aristocrat | N |
member-collection holonymy (has_member_holonym) | from members to their groups | worker —> working class | N |
portion-mass meronymy (has_portion_meronym) | from the whole mass to a portion | millennium —> century | N |
portion-mass holonymy (has_portion_holonym) | from a portion to the whole mass | month —> year | N |
causation (causes) | a verb causes a result state | destroy —> ruined put —> stand | V —> A V —> V |
active entailment | from verbs (events) to the verbs (events) they entail | light up —> shine | V |
passive entailment (is_entailed_by) [few occurrences] | from verbs (events) that follow from other verbs (events) | sleep—> snore | V |
association (is_related_to) | two concepts are associated with each other | soccer <—> goal | N A V |
Relation | Definition | Example | Classes |
---|---|---|---|
synonym | two lemmas with approximately the same meaning | diligent, hard-working | N A V |
antonym | two lemmas that are considered to be opposites | open <—> closed | N A V |
pertainymy | between a denominal adjective and its nominal base, where the noun forms the basis of the adjective and determines its meaning | financial —> finance | A —> N |
participle | from a verb to an adjective, where the adjective corresponds formally with the participle of a verb, but additionally shows idiosyncratic semantic features | limit —> limited | V —> A |
compound relations | semantic relation between compound and modifier | spiderweb is_product_of spider | N |
Compound Relations
Compound relations specify the semantic relation which can be established between a compound and its modifier. The inventory of those relations is based on the results of a project in the Collaborative Research Centre (SFB) 833 "The Construction of Meaning: The Dynamics and Adaptivity of Linguistic Structures" at the University of Tübingen: "Corpus-based Semantic Composition Models for Phrases". In this project a complex annotation scheme was developed to characterize the semantic relation between the modifier and the head of a given noun compound. We adapted this scheme and reformulated it as a relation between modifier and compound.
The following examples illustrate some possible relations of compounds with the head ‚Haus’:
Holzhaus (wooden house) [has_material] Holz (wood)
Gästehaus (guest house) [has_user] Gast (guest)
Autohaus (car dealership) [has_goods] Auto (car)
Baumhaus (tree house) [has_location] Baum (tree)
This table shows all compound relations with an example of each:
GN-relation | Example |
---|---|
has_active_usage | Schlafwagen has_active_usage Schlaf |
has_occasion | Geburtstagsgeschenk has_occasion Geburtstag |
has_attribute | Zauberland has_attribute Zauber |
has_appearance | Kugelfisch has_appearance Kugel |
has_construction_method | Blockhütte has_construction_method Block |
has_container | Dosenmilch has_container Dose |
is_container_for | Mokkatasse is_container_for Mokka |
has_consistency_of | Panzerglas has_consistency_of Panzer |
has_component | Chlorwassser has_component Chlor |
has_owner | Metzgerladen has_owner Metzger |
is_owner _of | Hauseigentümer is_owner _of Haus |
has_function | Grenzzaun has_function Grenze |
has_manner_of_functioning | Sonnenuhr has_manner_of_functioning Sonne |
has_origin | Ackersalat has_origin Acker |
has_production_method | Pfannkuchen has_production_method Pfanne |
has_content | Bilderbuch has_content Bild |
has_no_property | Maultier has_no_property Maul |
has_habitat | Bachforelle has_habitat Bach |
has_location | Almhütte has_location Alm |
is_location_of | Schlossberg is_location_of Schloss |
has_measure | Literflasche has_measure Liter |
is_measure_of | Blutzuckerspiegel is_measure_of Blutzucker |
has_material | Holzhaus has_material Holz |
has_member | Kinderchor has_member Kind |
is_member_of | Marinesoldat is_member_of Marine |
has_diet | Ameisenbär has_diet Ameise |
is_diet_of | Katzenfutter is_diet_of Katze |
has_eponym | Panflöte has_eponym Pan |
has_user | Taucherbrille has_user Taucher |
has_product | Autofabrik has_product Auto |
is_product_of | Spinnennetz is_product_ofSpinne |
has_prototypical_holder | Altartuch has_prototypical_holderAltar |
is_prototypical_holder_for | Kleiderbügelis_prototypical_holder_forKleid |
has_prototypical_place_of_ usage | Gartenbank has_prototypical _place_of_usage Garten |
has_relation | Bankdirektor has_relation Bank |
has_raw_product | Apfelsaft has_raw_product Apfel |
has_other_property | Jägerzaun has_other_property Jäger |
is_storage_for | Bildarchiv is_storage_for Bild |
has_specialization | Augenarzt has_specialization Auge |
has_part | Gelenkbus has_part Gelenk |
is_part_of | Autodach is_part_of Auto |
has_topic | Sportzeitung has_topic Sport |
is_caused_by | Regenbogen is_caused_by Regen |
is_cause_for | Schauerroman is_cause_for Schauer |
is_comparable_to | Satellitenstadt is_comparable_to Satellit |
has_usage | Handelsschiff has_usage Handel |
has_result_of_usage | Wärmflasche has_result_of_usage Wärme |
has_purpose_of_usage | Autoschlüssel has_purpose_of_usage Auto |
has_goods | Blumenladen has_goods Blume |
has_time | Abendzeitung has_time Abend |
is_access_to | Haustür is_access_to Haus |
has_ingredient | Obstkuchen has_ingredient Obst |
is_ingredient_of | Kaffeemilch is_ingredient_of Kaffee |
EuroWordNet Tests
The following relationship tests are taken from the EuroWordNet General Document.
Hyperonymy / Hyponymy:
EuroWordNet Hyperonymy / Hyponymy Test for Verbs
Verb synset X is hyponym of verb synset Y and verb synset Y is a hyperonym of verb synset X if the following test sentences can be answered accordingly:
X is Y + AdvP/AdjP/NP/PP. ('To run' is 'to go' fast.) | yes |
Y is X + AdvP/AdjP/NP/PP. ('To go' is 'to run' fast.) | no |
EuroWordNet Hyperonymy / Hyponymy Tests for Nouns
Note that this test can also be used for synonymy. Therefore a second test for species, kinds, races, and brands has been developed.
Test 1 (also for synonymy):
X is a hyponym/synonym of Y and Y is a hyperonym/synonym of X if the following test sentences apply:
An X is a Y with specific characteristics. (A car is a vehicle with specific characteristics.) It is an X and thus also a Y. (It is a car and thus also a vehicle.) If it is an X, it has to be also a Y. (If it is a car...) | yes |
Inversion of the above sentences (A vehicle is a car with specific characteristics.) (It is a vehicle and thus also a car.) (If it is a vehicle ...) | no |
Test 2 (for species, kinds, races, and brands):
Note that this test cannot be used for synonymy.
X is a kind/a type/a race/a species/a brand of Y. (A car is a kind of vehicle.) | yes |
Inversion of the above sentence (A vehicle is a kind of car.) | no |
Meronymy / Holonymy:
EuroWordNet Meronymy / Holonymy Test
X has a holonym Y and Y has a meronym X if
(an) X constitutes a part of (a) Y (a) Y has (an) X. | yes |
Inversion of the above sentences | no |
Synonymy:
The definition of synonymy is taken from (Miller et al., 1990): “two expressions are synonymous in a linguistic context C if the substitution of one for the other in C does not alter the truth value”.
EuroWordNet Synonymy Test for Nouns
if it is (a/an) X then it is also (a/an) Y
if it is (a/an) Y then it is also (a/an) X
Example:
if it is a fiddle then it is a violin
if it is a violin then it is a fiddle
synset variants {fiddle, violin}
EuroWordNet Synonymy Test for Verbs
If something/someone/it Xs then something/someone/it Ys
If something/someone/it Ys then something/someone/it Xs
If something/someone/it begins then something/someone/it starts
If something/someone/it starts then something/someone/it begins
synset variants: {begin, start}
Antonymy
EuroWordNet Antonymy Test for Verbs
In EuroWordNet antonymy is a relation between variants, i.e. between elements of the same synset: Synset variant X is an antonym of synset variant Y if X is the opposite of Y.
Test sentences:
If sth./sb. X-s, he/she/it does not Y. (If she borrows sth., she does not lend it.) | yes |
If sth./sb. Y-s, he/she/it does not X. (If she lends sth., she does not borrow it.) | yes |
The following conditions have to apply for this relation:
- X and Y share the same hyperonym, thus they are elements of a co-hyponym synset (this prevents, e. g., that verbs like eat and sleep are designated as antonyms).
- There is a hyperonym of X, which is the opposite of a hyperonym of Y.
- Both verbs involve the same participants, which play, however, different roles in the situations (i.e. states, events or processes) that are described by these verbs (example: give and receive are antonyms because the indirect object of give , i.e. the addressee, who is involved in the event, is subject of receive ).
EuroWordNet further assumes a relation called near_antonymy which applies to entire synsets, i.e. the antonymy relation holds between all members of the synset. Besides this, the test is the same as for antonymy (including the three conditions).
EuroWordNet Antonymy Test for Nouns
Antonymy and near_antonymy are also distinguished for nouns. The test is simpler than for verbs however:
X is an antonym of Y and Y is an antonym of X if the following test sentences apply:
X and Y are both a kind of Z, but X is the opposite of Y. (i.e. Z is a hyperonym of X and Y.) (Love and hate are both a kind of emotion, but love is the opposite of hate.) | yes |
Inversion of the above sentence (Hate and love are both a kind of emotion, but hate is the opposite of love.) | yes |
As for verbs, the condition that Z is a hyperonym of both X and Y is necessary in order to guarantee that the antonymy relation is stated in a reasonable, competitive denotational range.