12.4 File Change Semantics and update semantics

File Change Semantics

File change semantics models how sentences dynamically update our understanding of a discourse. Rather than treating sentence meaning as a static set of truth conditions, this framework treats each sentence as an instruction that changes the current information state. The central metaphor is a "file" containing "cards," where each card represents an entity that's been mentioned and stores what we know about it.

This approach is especially useful for explaining how indefinites introduce new entities, how definites refer back to established ones, and how phenomena like anaphora and presupposition work across multiple sentences.

Key Concepts

A file represents the current state of the discourse. Inside the file, each file card corresponds to a discourse referent, an entity that's been mentioned or implied. Each card stores the properties and relations we've accumulated about that referent so far.

• File states are snapshots of the file at a given point in the discourse. As each new sentence is processed, the file state changes.
• File updates are the changes a sentence triggers. A sentence might add a brand-new card (for a newly introduced entity) or write new information onto an existing card.

The key shift from classical semantics: a sentence's meaning isn't just its truth conditions. It also includes its context change potential, the way it transforms one file state into another.

Application in Discourse Interpretation

Because each sentence updates the file, interpretation is inherently sequential. The meaning of a sentence depends on the file state it receives as input, and it produces a new file state that the next sentence will use.

Consider a short discourse:

1. "A man walked in." This creates a new file card for the man (call it card $x$). The card records the property walked in.
2. "He sat down." The pronoun "he" is resolved to the existing card $x$. The card is then updated with the additional property sat down.

This is how file change semantics handles anaphora: pronouns don't have fixed referents on their own. They point back to cards already in the file. Presuppositions work similarly: a presupposition-triggering expression requires that certain information already be present in the file state. If it isn't, the update fails (or requires accommodation).

Treatment of Indefinites and Definites

The distinction between indefinites and definites maps neatly onto file operations:

• Indefinites (e.g., a man) introduce a new discourse referent. Processing an indefinite means creating a fresh file card and writing the descriptive content onto it.
• Definites (e.g., the man) refer to an existing discourse referent. Processing a definite means locating a unique card in the file that matches the descriptive content. If no such card exists, or if multiple cards match, the definite fails to refer properly.

This captures an intuitive asymmetry: indefinites expand the file by adding new entities, while definites maintain coherence by linking back to what's already established.

Comparison with Other Dynamic Semantic Theories

File Change Semantics vs. Update Semantics

Both file change semantics (Heim) and update semantics (Veltman) are dynamic theories that treat sentences as functions from input states to output states. The difference lies in what those states represent and what the theories are designed to explain.

• Update semantics models information states representing a speaker's knowledge or beliefs. Sentences are functions that map one information state to another. It's particularly well-suited for analyzing epistemic vocabulary like might and must, where the relevant question is what possibilities are still open.
• File change semantics models discourse referents and their accumulated properties. Its file-and-card structure gives a more fine-grained representation of which entities are in play and what we know about each one. This makes it especially strong for analyzing anaphora, definite descriptions, and donkey sentences.

Both frameworks agree that meaning is context-dependent and that sentences change the context for everything that follows. The choice between them often depends on the phenomenon you're trying to analyze: referent tracking favors file change semantics, while epistemic reasoning favors update semantics.