TL;DR
This paper investigates the often-overlooked argument structure of Russian adverbial derivatives (AdvDs), showing they inherit arguments from their base nouns but realize them in diverse and complex ways. The authors propose a novel, compact notation for representing these argument structures in dictionaries, crucial for advanced NLP tasks like semantic parsing and question answering.
What Problem Does This Paper Address?
Many natural language processing (NLP) applications struggle with understanding the full meaning of sentences because they don't adequately capture the argument structure of all word classes. While verbs and nouns have been extensively studied in this regard, adverbial derivatives (AdvDs) – words like "in a rage" or "with pleasure" derived from nouns – have been largely ignored. These AdvDs, despite their adverbial form, retain semantic connections to their source nouns and can govern arguments (e.g., "under the pretext of illness"). The challenge is that these arguments can appear in various positions and have complex relationships with the AdvD and the main verb of the sentence, making them difficult to identify and represent systematically. This lack of understanding hinders tasks that require deep semantic comprehension, such as recognizing textual entailments or extracting information.
Methods
The paper's core contribution lies in its linguistic analysis and proposed representational framework. The authors analyze Russian adverbial derivatives, categorizing them into grammatical (verbal adverbs) and lexical types. They meticulously examine how the "actants" (arguments) of the base noun are realized when the noun is transformed into an AdvD. The key innovation is the development of a compact notation for representing these argument structures within a dictionary. This notation uses an index attached to an "Adv" lexical function, where each position in the index corresponds to a valency slot of the keyword. The index specifies whether the slot can be filled, and if so, how (e.g., by the syntactic governor, a dependent of the governor, or the topic of the clause). This approach builds upon the "Meaning-Text" theory and its concept of lexical functions.
Datasets
Not discussed in the paper.
Key Insights and Ideas
- AdvDs are not just modifiers: Contrary to intuition, AdvDs are not simply adverbs; they actively govern arguments inherited from their base nouns, albeit in syntactically diverse ways.
- Argument realization is complex: The arguments of AdvDs can be blocked, filled by a dependent of the AdvD itself, attached to the dominating verb, or appear elsewhere in the clause, identified by syntactic position, semantic role, or communicative function.
- Semantic role is key: For certain AdvDs, like "v podarok" (as a present), semantic roles (Agent, Theme, Recipient) are more crucial than syntactic position for determining argument realization.
- Topic-sensitivity matters: Some AdvDs, like "v bol'šinstve" (mostly), are sensitive to the communicative structure of the sentence, with their arguments being filled by the "Topic."
- A unified representational approach is needed: The paper argues for a systematic way to represent AdvD argument structures, bridging the gap between their lexical origins and their syntactic realization.
Key Results
The paper provides a detailed linguistic analysis of various Russian adverbial derivatives, illustrating with numerous examples how their argument structures differ from their base nouns. It demonstrates the diverse ways arguments can be realized, including:
- Valence slots filled by a dependent of AdvD: Examples like "pod predlogom golovnoj boli" (under the pretext of a headache) show how the AdvD can govern its own dependent, with reversed governing properties compared to the base noun "predlog" (pretext).
- Valence slots filled by the dominating verb: Examples like "v dokazatel'stvo" (as proof) show the AdvD's argument being realized by the main verb's actant.
- Valence slots filled by dependents of the dominating verb: This category is further broken down by syntactic function, semantic role, and communicative function, showcasing nuanced relationships. For instance, "po privyčke" (by habit) is shown to be sensitive to the subject of the main verb, while "v podarok" (as a present) relies on semantic roles.
The paper proposes a notation that compactly represents these complex argument structures, aiming to make this information accessible for computational use.
Limitations and Future Work
The authors acknowledge that their work is a linguistic investigation and not directly about NLP implementation, though they emphasize its utility for NLP applications. They focus on Russian adverbial derivatives and suggest that similar analyses could be extended to other languages. The proposed notation is a step towards a more comprehensive lexical representation, but its practical integration into existing NLP systems would require further development. Future work could involve building a lexicon with these detailed argument structures and testing their effectiveness in downstream NLP tasks.
Summary updated