A Computational Treatment of Korean Temporal Markers, OE and DONGAN

Hyunjung SonACL 2003

TL;DR

This paper presents a computational framework for understanding Korean temporal markers "OE" and "DONGAN" using typed lambda calculus. By formalizing their semantic contributions, the authors demonstrate how these markers precisely specify event times, addressing ambiguities often left by Korean tense markers. This work contributes to more robust natural language processing for Korean by providing a compositional semantic representation.

What Problem Does This Paper Address?

Understanding the precise timing of events in natural language is crucial for accurate meaning representation. In Korean, while tense markers exist, they can be ambiguous, allowing for multiple interpretations of an event's temporal scope (e.g., simple past, completion, resultant state). This ambiguity makes it challenging to establish a coherent temporal interpretation of sentences. The paper specifically addresses how temporal adverbs like "OE" and "DONGAN" can resolve this ambiguity by providing more precise temporal information, acting as more reliable indicators of event time than tense/aspect alone.

Methods

The authors employ a computational approach based on typed lambda calculus to formalize the semantic properties of the temporal markers "OE" and "DONGAN." They define linguistic time as ordered, discrete, and composed of instants, extended intervals, and durations, each with a formal representation. The paper introduces a set of functional terms that capture these temporal notions and their relationships (e.g., beginning, ending, duration, anterior to, simultaneous with).

A key aspect of their method is the integration of these formalized temporal semantics into a computational parser built on Renaud's G Grammar. This grammar is designed for compositional semantic interpretation, meaning it builds the meaning of a sentence by combining the meanings of its parts. The process involves defining word meanings (dico) and composition rules that handle syntactic and semantic constraints, ultimately producing a semantic representation of the entire sentence.

Datasets

Not discussed in the paper. The paper mentions using "three thousand sentences" for analysis but does not specify their source or if they constitute a formal dataset.

Key Insights and Ideas

  • Temporal Adverbs as Primary Time Indicators: The paper strongly advocates for the view that temporal adverbs like "OE" and "DONGAN" are more reliable than tense/aspect markers for determining event time in Korean, due to the inherent ambiguity of the latter.
  • Formalizing Temporal Notions: The use of typed lambda calculus provides a rigorous way to define and manipulate concepts like instants, intervals, and durations, enabling precise semantic representation.
  • Compositional Semantics for Korean: By integrating temporal semantics into a G Grammar framework, the authors demonstrate a method for building sentence meaning compositionally, which is crucial for handling complex linguistic phenomena.
  • Distinguishing Event Types: The paper highlights the importance of distinguishing between mono-occurrent (single event) and multi-occurrent (multiple events) sentences, as temporal adverbs can impose different constraints depending on the event type.
  • OE vs. DONGAN Nuances: While both mark temporal aspects, "OE" can indicate the specific moment an event occurred or a period of recurrent events, whereas "DONGAN" often signifies the maximal duration of an event.

Key Results

The paper successfully demonstrates a computational method for representing the temporal semantics of Korean sentences involving "OE" and "DONGAN." Through examples and formalisms, it shows how these adverbs contribute to disambiguating event times. The authors' G Grammar-based parser is shown to construct semantic representations compositionally, handling various syntactic structures and semantic constraints associated with these temporal markers.

Limitations and Future Work

  • Future Work: The authors plan to investigate Korean time adverbials like "MANE" and the "zero particle," which are believed to signal telicity and are frequent in informal discourse.
  • Limitations: The paper acknowledges that the "zero particle" has a highly context-dependent meaning, which presents a challenge. While the paper presents a framework, the full implementation and evaluation of the Korean parser are not detailed, and the "datasets" used are not formally defined. The paper also notes that the distributional pattern of events can disappear when the nuclear sentence is modified by quantification, suggesting further complexity to explore.

Summary updated

BibTeX
@inproceedings{son2003computational,
  author = {Hyunjung Son},
  title = {{A Computational Treatment of Korean Temporal Markers, OE and DONGAN}},
  booktitle = {Annual Meeting of the Association for Computational Linguistics},
  year = {2003},
}

Loading related content...

Figure