TL;DR
This paper investigates German "light verb constructions" (LVCs) like "unter Beobachtung stehen" (to be under observation). It reveals that these LVCs, despite similar grammatical structures, fall into distinct "families" based on their interpretation patterns, which are driven by the semantic type of the noun within the prepositional phrase. This work offers a systematic corpus-based approach to understanding LVC families, a concept previously underexplored in German linguistics.
What Problem Does This Paper Address?
The paper tackles the challenge of understanding the semantic variability within seemingly similar grammatical constructions. Specifically, it focuses on German light verb constructions (LVCs) of the form "stehen unter NP" (to stand under NP). While LVCs are known to involve a semantically "light" verb and a phrasal element, their precise meaning can vary significantly. The authors highlight that LVCs often form "families" that share systematic interpretation patterns. However, the exact criteria for identifying these families and the properties that define them have not been systematically studied, especially in German. This paper aims to fill this gap by conducting a corpus study to identify and characterize these LVC families.
Methods
The study employs a multi-step corpus annotation process.
- First Annotation Step: Sentences were screened to identify instances where the "unter NP" phrase functions as a syntactic complement to the verb "stehen." This was determined by checking if the phrase could be omitted without affecting sentence acceptability.
- Second Annotation Step: The remaining sentences were classified into "heavy" and "non-heavy" uses of "stehen." Heavy uses involve "stehen" in its literal posture or locational sense, substitutable by other posture verbs. Non-heavy uses encompass light and abstract uses where such substitution is not possible.
- Third Annotation Step: For the non-heavy uses, the study focused on identifying LVC families by checking if the construction could be paraphrased by either an "event passive" construction (e.g., "beobachtet werden" - to be observed) or a "state passive" construction (e.g., "geschockt sein" - to be shocked). This step was crucial for distinguishing between different interpretation patterns.
The semantic type of the PP-internal nouns was also classified using criteria such as their ability to be subjects of verbs like "geschehen" (to happen) or "beenden" (to end), distinguishing between event-denoting and state-denoting nouns.
Datasets
The study utilized the Tagged-C2 archive of the German reference corpus (DeReKo). This corpus is described as containing newspaper articles and comprising over 1 billion words across more than 4.4 million texts. A subset of sentences was randomly collected for manual annotation. The search criterion used was "&stehen \s0 unter", yielding 80,255 hits, from which 8,023 sentences were randomly selected for annotation. 55 of these were excluded due to incompleteness.
Key Insights and Ideas
- LVCs form "families": Despite similar surface forms (e.g., "stehen unter NP"), LVCs can exhibit distinct interpretation patterns, forming families that share these patterns.
- Semantic type of the NP is key: The core driver for these different interpretation patterns is the semantic type of the noun within the "unter NP" phrase. Event-denoting nouns lead to "event passive" families, while state-denoting nouns lead to "state passive" families.
- "stehen unter NP" is not always a light verb construction: The initial filtering step identified that not all instances of "stehen unter NP" involve the PP as a syntactic complement, meaning some are not LVCs.
- "stehen" in LVCs is not semantically empty: While "light," the verb "stehen" contributes to the event structure and often embeds the passive-like meaning of the PP into a state predication.
- Preposition "unter" plays a crucial role: The passive-like interpretation is strongly linked to the specific use of the preposition "unter," independent of the light verb itself.
- Artefact nouns can also form state passive families: Nouns denoting artefacts (like "Drogen" - drugs) can be associated with a specific state (e.g., being under the influence) and thus contribute to state passive readings.
Key Results
- The study identified two main LVC families for "stehen unter NP": the event passive-family and the state passive-family.
- The event passive-family is represented by 33 different LVCs, while the state passive-family has 19 members.
- Out of the analyzed non-heavy "stehen unter" constructions, 1,335 occurrences were paraphrased by an event passive construction, and 1,524 by a state passive construction. These two families together account for 49.23% of all non-heavy uses of "stehen unter" in the analyzed sample.
- A preference analysis of causative light verbs (stellen/setzen) showed that the event passive-family prefers "stellen" (to put/cause to stand), while the state passive-family prefers "setzen" (to put/cause to sit). For example, "unter Beobachtung stellen" (to put under observation) occurred 201 times, while "unter Schutz stellen" (to put under protection) occurred 2179 times, compared to "unter Schock setzen" (1 vs. 7) and "unter Stress setzen" (1 vs. 244).
Limitations and Future Work
- Limited Scope: The study focused specifically on "stehen unter NP" LVCs. The authors acknowledge that other LVCs and other languages might exhibit different family structures.
- Future Work:
- Investigating LVC families in other languages, particularly those with light verbs meaning "stand" (like Dutch).
- Exploring further characteristics that differentiate LVC families, such as causativization preferences (as hinted at in Table 4).
- Developing automated methods for identifying "stehen-LVCs" by training learning algorithms on the annotated data.
Summary updated