A Model of Natural Language Processing of Time-Related Expressions

Yutaka KusanagiCOLING 1980

TL;DR

This paper introduces a novel computational model for processing time-related expressions in natural language. By incorporating the concept of "focus" – the speaker's attentional spotlight on a specific time period – the model bridges the gap between real-world situations and linguistic representations, paving the way for more nuanced language understanding.

What Problem Does This Paper Address?

Natural language is notoriously flexible when it comes to expressing time. A single real-world event can be described using various grammatical structures (e.g., different tenses and aspects), and conversely, the same linguistic expression can refer to slightly different temporal nuances. This lack of a one-to-one correspondence between real-world time and linguistic time poses a significant challenge for computers trying to understand and generate language. The authors argue that previous models have struggled because they haven't adequately accounted for the speaker's subjective perspective. This paper aims to solve this by proposing a model that explicitly incorporates the speaker's cognitive "focus" on time.

Methods

The proposed model operates on three distinct levels:

  1. Extra-linguistic Level: This level represents the "real world" and stores information about situations, including their start and end times. Situations are categorized as either dynamic (happening, like "reading") or static (existing, like "being calm"). Dynamic situations can be momentary ("a light going on") or enduring ("reading a book").
  2. Cognition Level: This is the core innovation. It introduces the concept of "focus," defined as the specific time period the speaker directs their attention to. This focus is then matched against the extra-linguistic situations. The way the focus aligns with or diverges from the situation's temporal boundaries is crucial for determining linguistic features.
  3. Linguistic Level: This level uses syntactic, morphological, and phonological rules to construct the actual sentences based on the decisions made in the cognition level.

The model uses a SNOBOL programming language implementation. The paper details how aspect categories (e.g., perfective, imperfective, stative) are determined by the relationship between the situation's temporal points (start, end) and the focus's temporal points (start, end). Tense (past, present, future) is then assigned based on the relationship between the focused time and the time of utterance.

Datasets

Not discussed in the paper. The paper focuses on the model's architecture and logic, using example sentences and hypothetical situations for illustration rather than formal datasets.

Key Insights and Ideas

  • The Power of "Focus": The central idea is that understanding time in language requires more than just knowing when an event happened; it's about where the speaker is looking in time. This cognitive focus is the key to resolving ambiguities.
  • Bridging the Gap: The three-level model (extra-linguistic, cognition, linguistic) provides a structured way to connect objective reality with subjective linguistic expression.
  • Aspect as a Cognitive Tool: The paper demonstrates how aspectual categories are not just grammatical markers but reflect how a speaker cognitively frames the temporal structure of an event.
  • Language-Specific Rules: While the core model aims for universality, the paper acknowledges that the final linguistic level requires language-specific rules to handle the nuances of morpheme assignment and sentence construction.

Key Results

The paper presents a conceptual model and its implementation logic, rather than quantitative results from a large-scale evaluation. The "results" are demonstrated through:

  • Example Sentence Analysis: The paper shows how the model can correctly assign aspects and tenses to example sentences, illustrating the logic of the focus-based approach. For instance, it explains how different English sentences describing the same past event (e.g., "I wrote a letter" vs. "I was writing a letter") are distinguished by the speaker's focus.
  • Programmatic Logic: The appendices provide detailed SNOBOL code and diagrams that illustrate the step-by-step determination of aspect and tense based on temporal relationships and focus.

Limitations and Future Work

  • Language Specificity: The authors acknowledge that while the model's core is universal, the linguistic level requires language-specific programming. The current model is primarily demonstrated with English and Japanese examples.
  • Complexity of Expressions: The paper notes that expressions of habit, experience, and atemporal situations, as well as relative tenses, were omitted for simplicity and would require further elaboration.
  • Future Work: The authors suggest that further research should focus on expanding the morpheme assignment rules for various natural languages to identify more universal rules and minimize language-peculiar ones. They also propose that the time representation system could be expanded to include seconds, milliseconds, and years.

Summary updated

BibTeX
@inproceedings{kusanagi1980model,
  author = {Yutaka Kusanagi},
  title = {{A Model of Natural Language Processing of Time-Related Expressions}},
  booktitle = {International Conference on Computational Linguistics},
  year = {1980},
}
Figure