Adposition Supersenses v2

Nathan Schneider, Jena D. Hwang, Archna Bhatia, Na-Rae Han, Vivek Srikumar, Tim O'Gorman, Omri AbendarXiv 2017 · arXiv

Abstract

This paper presents SNACS v2.6 (Semantic Network of Adposition and Case Supersenses), a comprehensive framework for labeling the semantic roles of prepositions and case markers. By introducing a dual-layer "construal" system and a hierarchical taxonomy of 52 labels, the authors provide a standardized method for mapping spatial and causal language to abstract concepts. This version refines previous iterations by simplifying the hierarchy and integrating possessives and infinitives into a unified semantic system.

Introduction

In Natural Language Processing (NLP), adpositions (e.g., in, at, on, for) present a significant challenge due to their extreme polysemy. A single word like "in" can denote a physical location ("in the box"), a temporal point ("in the morning"), or a state ("in a hurry"). Existing semantic labeling schemes often struggle with being either too granular for consistent annotation or too coarse for meaningful machine learning applications. The SNACS project addresses this by establishing a "Goldilocks" zone: a linguistically grounded inventory of 52 "supersenses" broad enough for cross-linguistic use and large-scale dataset annotation.

Methodology: The SNACS Hierarchy

The core of the SNACS v2.6 framework is a hierarchical taxonomy organized into three major subhierarchies: Circumstance, Participant, and Configuration.

  1. Circumstance: Covers space, time, and non-core event properties (e.g., TIME, LOCUS, PURPOSE).
  2. Participant: Describes entities playing specific roles in an event (e.g., AGENT, THEME, RECIPIENT).
  3. Configuration: Defines static relationships between entities or properties (e.g., POSSESSOR, IDENTITY, PARTPORTION).

A key innovation in v2.6 is the removal of "multiple inheritance," which previously allowed a label to belong to two parents. This cleanup ensures the network is no longer "tangled," making it significantly easier for computational models to navigate.

The Construal Mechanism

Beyond the labels themselves, SNACS introduces a "construal" mechanism to capture how humans use metaphors. Every adposition is assigned two supersenses:

  • Scene Role: The underlying semantic relation in the described event.
  • Function: The meaning suggested by the specific lexical choice (the "surface" meaning).

For example, in the phrase "Put the book on the shelf," the Scene Role is GOAL (the destination), while the Function is LOCUS (the static location typically associated with "on"). This is formally notated as GOAL ⇝ LOCUS. This system allows the framework to explain abstract metaphors, such as "falling into a depression" (STATE ⇝ GOAL), where a motion word describes a change of state.

Datasets and Applications

The SNACS framework has been applied to several major linguistic resources:

  • STREUSLE Corpus: A dataset of web reviews (v4.5) annotated using the v2.6 guidelines.
  • Xposition: An online multilingual database (xposition.org) that allows researchers to browse supersenses alongside real-world corpus data.
  • Cross-linguistic Validation: The inventory was refined through the analysis of English, Hebrew, Hindi, Korean, and German to ensure its potential as a universal semantic standard.

Discussion and Limitations

While SNACS v2.6 provides a stable, 114-page manual for annotation, challenges remain. The current guidelines are heavily tailored to English, though the underlying inventory aims for universality. Future work involves publishing specific documentation for other languages to account for diverse grammatical structures, such as case markers in Turkish or Hindi. Additionally, "idiomatic" expressions (e.g., "out of business") continue to pose boundary-case challenges for annotators.

Conclusion

SNACS v2.6 represents a significant step toward machines that truly "understand" the relational glue of language. By balancing linguistic depth with computational feasibility, the framework enables the development of automatic semantic taggers and improves cross-lingual information retrieval. The ultimate goal is the continued expansion of a multilingual database of adpositional semantics.

Summary updated

Abstract

This document offers a detailed linguistic description of SNACS (Semantic Network of Adposition and Case Supersenses; Schneider et al., 2018), an inventory of 52 semantic labels ("supersenses") that characterize the use of adpositions and case markers at a somewhat coarse level of granularity, as demonstrated in the STREUSLE corpus (https://github.com/nert-nlp/streusle/ ; version 4.5 tracks guidelines version 2.6). Though the SNACS inventory aspires to be universal, this document is specific to English; documentation for other languages will be published separately. Version 2 is a revision of the supersense inventory proposed for English by Schneider et al. (2015, 2016) (henceforth "v1"), which in turn was based on previous schemes. The present inventory was developed after extensive review of the v1 corpus annotations for English, plus previously unanalyzed genitive case possessives (Blodgett and Schneider, 2018), as well as consideration of adposition and case phenomena in Hebrew, Hindi, Korean, and German. Hwang et al. (2017) present the theoretical underpinnings of the v2 scheme. Schneider et al. (2018) summarize the scheme, its application to English corpus data, and an automatic disambiguation task. Liu et al. (2021) offer an English Lexical Semantic Recognition tagger that includes SNACS labels in its output. This documentation can also be browsed alongside corpus data on the Xposition website (Gessler et al., 2022): http://www.xposition.org/
BibTeX
@misc{schneider2017adposition,
  author = {Nathan Schneider and Jena D. Hwang and Archna Bhatia and Na-Rae Han and Vivek Srikumar and Tim O'Gorman and Omri Abend},
  title = {{Adposition Supersenses v2}},
  year = {2017},
  eprint = {1704.02134},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1704.02134},
}
Figure