Compounds in GermaNet

What makes compound splitting for German a challenging task is the fact that compounding is not always simple string concatenation, but often involves the presence of intervening linking elements or the elision of word-final characters in the modifier constituent of a compound (Henrich & Hinrichs, 2011). In GermaNet, all modifiers are lemmatized and if a modifier is ambiguous with respect to its word class (due to conversion), both possibilities are specified:

• Laufschuhe: lauf- (en) [verb] and (der) Lauf [noun]
• Baustelle: bau- (en) [verb] and (der) Bau [noun]

Compound splitting in GermaNet is supported by an automatic algorithm, which combines several individual compound splitters. Please see the referenced paper below for more information on the automatic splitting. All automatically split compounds are manually post-corrected and enriched with relevant properties before they are inserted into GermaNet.

Properties

The following properties are specified for modifiers and/or heads:

Abbreviation

If one part of the compound is an abbreviation, it is labelled as Abkürzung.

Examples:

Affixoid

Affixoids are morphemes with a special status between bound and free morphemes. As they have a clearly assigned meaning, it makes sense to split the respective words. The bound morpheme is labelled as Affixoid.

Examples:

Foreign Word

If one part (or more) of the compound is not a German word, it is labelled as Fremdwort. Note that those constituents which are borrowed words but are nowadays used as loanwords defined in a standard German dictionary (such as Duden) are not considered as foreign words in GermaNet (e.g. Drink and Pool in the examples below).

Examples:

Konfix

The label Konfix refers to a word which is borrowed from a foreign language, in many cases from Latin or Greek, and whose meaning stems from that particular language. Konfixes are bound morphemes, but in opposition to all other affixes two Konfixes can be combined to form a so-called Konfixkompositum. Those Konfixkomposita are not split in GermaNet, whereas compounds existing of a Konfix and a native word are split.

Examples:

Opaque Morpheme

Modifiers whose meaning is not transparent any more without considering the etymology of the word are labelled with the property opaques Morphem.

Examples:

Proper Name

If the whole compound is a named entity, it is not split in GermaNet. If only the modifier is a proper name, the compound is split and the label Eigenname is added to the modifier.

Examples:

Virtual Word Form

Virtual word forms, labelled as Virtuelle Bildung, are regularly built according to existing word formation rules. However, they do not exist in isolation, but only as part of a compound.

Examples:

Word Group

Modifiers consisting of a phrase are marked as Wortgruppe and the parts of the phrase are annotated as the modifier.

Examples:

The following table gives an overview of the constituent parts of a compound (i.e. modifier and head) and the corresponding properties that are annotated for each constituent in GermaNet:

Disambiguation of Compound Components

Since the semantic interpretation of compounds typically depends on the meanings of their constituent elements, the availability of detailed information about these components is of considerable analytical value for many applications. However, the components of compounds are often polysemous, which makes adequate computational analysis difficult unless the relevant sense of each component is taken into account.
As a result, the polysemous constituents of the compounds contained in GermaNet have been systematically disambiguated, and all compound components have been annotated with the IDs of the corresponding lexical units.

Modifier Disambiguation

The link between a modifier and its respective sense is established through the specification of the corresponding ID. The semantic relationship between a compound and its modifier allows for a wide range of possible interpretations.

Examples:

If a compound contains two different modifiers, disambiguation is carried out for both components.

Example:

A modifier can sometimes be interpreted in multiple ways; in such cases, all potential meaning variants are recorded by specifying the relevant IDs.

Examples:

If the modifier is an affixoid, confix, foreign word, semantically opaque morpheme, a word class not included in GermaNet, or a complex word group, a semantic assignment is not possible.

Head Disambiguation

In most cases, a compound appears within the conceptual hierarchy as a direct or indirect hyponym of a higher-level hypernym. Accordingly, the head constituent is assigned the ID of the hypernym, since the compound semantically represents a subcategory of that hypernym.

Examples:

If the compound has a different hypernym than the head constituent, the head’s ID is assigned when the compound can semantically be interpreted as a type of that head.

Examples:

A Backform is a kind of Form (as an artifact), and a Surfbrett is a kind of Brett; however, their respective hypernyms in GermaNet are kitchenware for Backform and winter sports equipment for Surfbrett.

If the compound cannot be semantically interpreted as a kind of its head, no ID is assigned to the head constituent. For example, a Nichtraucher (“non-smoker”) is not a type of Raucher (“smoker”), and Acrylglas (“acrylic glass”) is not a type of Glas (“glass”).

Similarly, no ID is assigned if there is a part–whole relation between the compound and the head constituent.
In such cases, the relation to the head is recorded explicitly as a part–whole relation. For example, Viertelliter (“quarter liter”) is not a type of Liter (“liter”), but part of a liter, so no head ID is given. Instead, the part–whole relation is recorded as:
Viertelliter – has_portion_holonym – Liter

Figurative Meanings

If a compound is used idiomatically or metaphorically in its overall meaning, no semantic sense is assigned to either constituent.

Examples:
Frauenschuh (as an orchid species),  Eselsbrücke ("mnemonic"), Fettnäpfchen ("social blunder")

If only the head constituent is used metaphorically, no ID is assigned to the head, but the modifier is still disambiguated.

Examples:

Download

In addition to the information described above that is included in GermaNet (since release 8.0), a list of split compounds with their modifier(s) and head is freely available for download here:

• GermaNet v20.0 (2025): list of 130901 split nominal compounds
• GermaNet v19.0 (2024): list of 126733 split nominal compounds
• GermaNet v18.0 (2023): list of 121655 split nominal compounds
• GermaNet v17.0 (2022): list of 115366 split nominal compounds (Updated June 28, 2022, with corrections)
• GermaNet v16.0 (2021): list of 106780 split nominal compounds
• GermaNet v15.0 (2020): list of 98905 split nominal compounds (Updated June 15, 2020, with corrections)
• GermaNet v14.0 (2019): list of 91106 split nominal compounds
• GermaNet v13.0 (2018): list of 82309 split nominal compounds
• GermaNet v12.0 (2017): list of 74990 split nominal compounds
• GermaNet v11.0 (2016): list of 66059 split nominal compounds (v11.0: Updated Feb. 16, 2017)
• GermaNet v10.0 (2015): list of 54569 split nominal compounds
• GermaNet v9.0 (2014): list of 54759 split nominal compounds
• GermaNet v8.0 (2013): list of 40437 split nominal compounds
  (v8.0: Original list released on May 31, 2013, updated on June 12, 2013.)

The list of compound data is free for academic research as defined in GermaNet's academic research licence agreement. For any other intended purposes, please contact us.

The format of these split compounds is one compound per line: first the compound itself, then a <tab> space, then the modifier (in case of two modifiers, these are separated by the pipe (|) symbol), then a <tab> space again, and finally the head. For example:

Apfelbaum      Apfel   Baum
Goldmünze     Gold   Münze
Laufband       laufen|Lauf     Band

Reference

The following paper describes the automatic compound splitting that is performed before the manual post-correction. If you want to use the split compounds in the context of scientific or research work, please refer to the paper:

Verena Henrich and Erhard Hinrichs: Determining Immediate Constituents of Compounds in GermaNet. In Proceedings of Recent Advances in Natural Language Processing (RANLP 2011), Hissar, Bulgaria, September 2011, pp. 420-426.