Subcategorisation Lexicon for German Verbs

In lexical semantics and computational linguistics, a subcategorisation lexicon enumerates the possible argument structures that a verb can combine with. For German, subcategorisation is crucial because verbargument patterns interact with word order, case marking, and modal particles. This page summarises the main types of subcategorisation frames for German verbs, illustrates typical examples, and outlines how such a lexicon can be built and used.

1. Why Subcategorisation Matters for German

• Free word order: The position of objects (accusative vs. dative) can vary, but the case system still signals grammatical roles.
• Verbsecond (V2) and verbfinal clauses: Knowing whether a verb expects an object helps parsers predict clause boundaries.
• Complex predicates: Verbs combined with prepositions, particles, or infinitival complements require detailed frame information.
• Lexical ambiguity: Many German verbs are polysemous (e.g., fahren to drive vs. to travel). Different senses often have different subcategorisation patterns.

2. Basic Frame Types

Below are the most frequent subcategorisation frames found in German verb lexicons. The notation follows the tradition of Lezi (valency) where each argument is represented by its case or clause type.

Frame	Description	Typical Example
V	Intransitive verb, no verbal complement.	Er schlft.
V+NP[ACC]	Transitive verb taking a direct object.	Sie liest das Buch.
V+NP[DAT]	Verb that selects a dative argument (often a recipient or experiencer).	Er hilft dem Kind.
V+NP[ACC]+NP[DAT]	Ditranstive pattern accusative object (Theme) and dative object (Recipient).	Wir geben dem Mann das Buch.
V+PP[prep]	Verb selecting a prepositional phrase; the preposition is lexicalized.	Sie wartet auf den Zug.
V+S[INF]	Verb taking an infinitival complement (e.g., modal, perception).	Er versucht zu schlafen.
V+S[FIN]	Verb that selects a finite subordinate clause.	Ich glaube, dass er kommt.
V+NP[ACC]+S[INF]	Verb that takes both an object and an infinitival complement.	Sie lsst den Hund laufen.

3. Common Subcategories in Detail

3.1 Intransitives (V)

Pure intransitives do not assign a case to any NP. Many are motion or state verbs: schlafen, sterben, kommen. Some allow optional adverbials, but these are not part of the core valency.

3.2 Transitives (V+NP[ACC])

These verbs require an accusative object. The object can be a concrete noun phrase or a pronoun. The verb may also allow additional complements (e.g., schreiben + zu-PP). Example:

3.3 Dativeonly verbs

Verbs such as helfen, danken, glauben select a dative argument without an accusative counterpart. Note that some can also take an accusative object when a preposition is added (helfen bei).

3.4 Ditranstives (V+NP[ACC]+NP[DAT])

Classic ditransitive patterns include geben, schenken, zeigen. Word order is flexible: the dative may precede the accusative, especially in written German.

3.5 Prepositional complements (V+PP[prep])

Some verbs embed a specific preposition, and the case of the PP is fixed by the preposition. Examples: abhngen von, teilnehmen an, warten auf. The preposition is part of the lexical entry, not a free argument.

3.6 Infinitival complements (V+S[INF])

Modal verbs (knnen, mssen), perception verbs (sehen, hren), and causative verbs (lassen, bringen) take a bare infinitive. The infinitive often inherits the case of a missing object (e.g., Ich sehe den Mann laufen).

3.7 Finite subordinate clauses (V+S[FIN])

Complementizers are typically dass, ob, or a question word. Verbs that select such clauses include glauben, behaupten, fragen.

4. Building a Subcategorisation Lexicon

4.1 Data sources

• Parsed corpora (e.g., Tiger, Hamburg Dependency Treebank)
• Lexical databases (GermaNet, VerbNetGerman)
• Manual dictionaries with valency information

4.2 Extraction pipeline

1. Tokenisation & POStagging: Use tools such as spaCy or TreeTagger.
2. Dependency parsing: Identify subjects (nsubj), objects (obj), indirect objects (iobj), and prepositional complements (obl).
3. Frame induction: Group observed argument configurations per lemmasense.
4. Frequency filtering: Discard frames that occur below a threshold to reduce noise.
5. Sense linking: Align frames with WordNet or OntoLex senses for multilingual applications.

4.3 Representation format

Lexicons are often stored in XML (e.g., TEI), JSONLD, or CSV. A simple JSON example for the verb geben:

{  "lemma": "geben",  "sense": "to give",  "frames": [    {"pattern": ["NP[DAT]", "NP[ACC]"], "frequency": 842},    {"pattern": ["NP[ACC]"], "frequency": 57}  ]}

5. Applications

• Parsing and generation: Knowing which arguments are obligatory guides syntactic parsers and NLG systems.
• Machine translation: Correctly mapping German case patterns to languages without case (e.g., English) relies on valency information.
• Semantic role labeling: Subcategorisation frames provide a scaffold for assigning roles like Agent, Patient, Recipient.
• Lexicography: Dictionary entries can display possible complements automatically.

6. Challenges Specific to German

6.1 Optionality and particle verbs

Many verbs combine with separable particles (aufstehen, zurckkommen) that can appear in different clause positions, affecting surface order without changing the underlying valency.

6.2 Reflexive constructions

Verbs like sich erinnern use a reflexive pronoun that occupies the accusative slot. The lexicon must treat the reflexive element as part of the frame.

6.3 Nominalisations

When a verb is nominalised (e.g., die Entscheidung), its argument structure is often expressed via genitive or prepositional complements, requiring crosscategory mapping.

7. Sample Entry for a Polysense Verb

Consider the verb fahren (to drive / to travel). It has two major senses with distinct frames:

• S1 (operate a vehicle): V+NP[ACC] (object = vehicle) e.g., Er fhrt das Auto.
• S2 (travel by means of transport): V+PP[mit]+NP[DAT] (instrument) e.g., Wir fahren mit dem Zug.

Encoding separate sense IDs prevents the two frames from being conflated.

8. Future Directions

Research continues on richer representations such as construction grammar where the interaction between verb, particles, and word order is modelled jointly. Neural models that integrate valency embeddings show promise for improving lowresource German parsing.

By maintaining a detailed, sensedisambiguated subcategorisation lexicon, researchers and developers can more accurately capture the syntactic and semantic behaviour of German verbs, leading to better linguistic tools and applications.