TL;DR
This paper proposes a novel computational approach to understanding how temporal adverbs like "for" and "in" can subtly change the meaning of verbs, a phenomenon known as adverbial aspect shift. By treating temporal prepositions as dynamic "presuppositions" and employing a supervaluational logic, the authors present an algorithm that derives multiple possible interpretations of sentences, offering a more cognitively plausible explanation for this linguistic flexibility.
What Problem Does This Paper Address?
In linguistics, verbs are often categorized by their inherent temporal properties, such as whether they describe a state (e.g., "know"), an activity (e.g., "run"), an accomplishment (e.g., "build a tower"), or an achievement (e.g., "arrive"). Temporal adverbs, like "for an hour" or "in an hour," can interact with these verbs to create different meanings. For instance, "run for an hour" describes the duration of the activity, while "arrive in an hour" specifies when the event will be completed.
The challenge arises when adverbs seem to modify verbs in ways that contradict their inherent aspectual type, leading to what's called "aspectual shift." Traditional approaches often struggle to explain this flexibility, sometimes resorting to ad-hoc rules or treating these shifts as exceptions. This paper tackles the problem of formally accounting for these shifts in a compositional and systematic way, aiming to explain how and why these meaning changes occur within a computational framework.
Methods
The core of the proposed method lies in two key ideas:
- Supervaluational Underspecification: Instead of assigning a single, definitive truth value (true or false) to a proposition, this approach allows for "truth-value gaps." This means that when there's ambiguity or conflict, the system doesn't commit to a single interpretation but acknowledges a range of possibilities. This is achieved by considering all possible "completions" of a partial model.
- Dynamic, Structural Presuppositions: Temporal prepositions like "in" and "for" are not treated as static modifiers but as dynamic elements that impose constraints on the verb's aspectual type. When these constraints conflict with the verb's inherent type, instead of causing outright ungrammaticality, they introduce a truth-value gap. This gap then triggers a "repair" mechanism.
The algorithm operates in three steps:
- Superposition: The meaning of the temporal adverbial (e.g., "for an hour") is combined with the verb's aspectual characterization.
- Reparation: If there's an aspectual clash (e.g., "arrive for an hour"), a truth-value gap is introduced for the conflicting aspectual feature.
- Determination: The truth-value gap is then resolved by exploring all possible ways to fill it, generating the different valid interpretations (aspectual shifts).
This approach is framed within Fernando's finite-state temporal semantics, which represents situations as regular languages accepted by finite-state automata, allowing for a detailed internal structure of events.
Datasets
Not discussed in the paper. The paper focuses on linguistic examples and theoretical formalization rather than empirical evaluation on large datasets.
Key Insights and Ideas
- Aspectual Shifts as Internal Restructuring: The paper argues that aspectual shifts are not simple mappings from one event type to another but rather internal restructurings of a situational concept. This means the meaning change happens "from inside" the representation of the situation.
- Prepositions as Dynamic Presuppositions: Treating prepositions as dynamic presuppositions that can introduce underspecification (truth-value gaps) is a crucial departure from static approaches. This allows for a more flexible handling of conflicts.
- Cognitive Plausibility: The proposed mechanism of introducing a gap and then exploring all possible ways to fill it is presented as a more cognitively plausible model of how humans might resolve ambiguity when encountering aspectual shifts.
- Language Internal Mechanism: The authors emphasize that the entire process of aspectual shifting is explained as a language-internal semantic mechanism, rather than relying on external world knowledge or pragmatic inference alone.
- Reciprocal Adaptations: The model allows for reciprocal adaptations between the adverbial and the verb, meaning both can influence the final interpretation, rather than just the adverbial "forcing" a change on the verb.
Key Results
The paper demonstrates how its algorithm can derive multiple, distinct readings for sentences involving aspectual shifts, such as:
- For telic verbs modified by "for" (e.g., "build a tower for an hour"), the algorithm derives readings corresponding to:
- The duration of the result state (e.g., the tower exists for an hour).
- The duration of the preparatory phase (e.g., the process leading up to building the tower takes an hour).
- An iterative reading (multiple instances of building a tower, each taking some time, within the hour).
- For atelic verbs modified by "in" (e.g., "swim in an hour"), the algorithm derives readings corresponding to:
- The pre-phase of the activity (e.g., the time before swimming starts).
- The duration of the entire situation (e.g., the swimming activity itself, bounded by the hour).
The paper provides formalizations for these examples using its finite-state automata framework.
Limitations and Future Work
- Filtering Interpretations: While the algorithm generates all possible readings, the paper acknowledges that further work is needed to filter these possibilities using contextual information and world knowledge to arrive at the single "best" interpretation in a given discourse.
- Pragmatic Principles: For certain cases, like the result state entailment for verbs like "read," the authors suggest that pragmatic principles might be necessary to fully account for the interpretation, beyond purely semantic mechanisms.
- Generalizability: While the paper focuses on "for" and "in" with specific verb types, future work could explore the broader applicability of the algorithm to other prepositions and aspectual categories.
- Empirical Validation: The paper is primarily theoretical and formal. Future work could involve empirical validation of the proposed model against linguistic data and human judgments.
Summary updated