In everyday conversation, pointing to an object while saying "take a look at this" is an effortless way to establish shared understanding. The physical environment acts as a common canvas, allowing speakers to anchor language to tangible things using spatial and temporal cues.
When language moves to the printed page, however, this physical canvas vanishes. Written text is inherently decontextualized. Without a shared physical world, words like this, that, here, and now can no longer rely on external sensory input. Instead, the text must manufacture its own environment, forcing the reader to actively build internal models of the objects, spaces, and concepts being discussed.
The Contextual Gap Between Speech and Text
In face-to-face communication, listeners benefit from immediate situational feedback, gestures, and a shared physical environment. Written language strips these away. As explored in From Conversations to Stories: An Analysis of Differences Between Oral and Written Language Comprehension, the transition from spoken dialogue to written narrative requires a profound cognitive shift. Instead of designating existing physical objects, text demands the mental construction of referents.
Rubin et al. argue that when a reader encounters a descriptive noun phrase (such as "the blue chair") or a deictic term (like "here"), they cannot look around the room to find its meaning. They must construct an internal coordinate system for the story world. If a reader mistakenly applies conversational heuristics—expecting "here" to mean the physical room they are sitting in rather than the fictional setting of the narrative—comprehension breaks down.
This challenge is echoed in early computational and developmental linguistics. The author of D. D. McDonald (ijcai 1977) notes that textual designations function analogously to spoken deictic expressions but carry an extra layer of complexity: they require the reader to generate entities and temporal points entirely from linguistic descriptions. When children learn to read, over-relying on the implicit, shared context of speech can hinder their ability to process decontextualized text. As argued by the authors of BruceLanguage Generation: The Linguistics Component, interpreting terms like this or there becomes significantly more difficult in text because the reader must establish a reference point purely from the surrounding narrative context.
Oral Communication:
[Speaker & Listener] ---> Shared Physical Space ---> Physical Object ("Look at this")
Written Communication:
[Reader] ---> Textual Context ---> Mental Construction ---> Virtual Referent ("Inside the room, this...")
Building Local Centers Within Discourse
How exactly does text signal the creation of these mental entities? It often happens through subtle shifts in pronoun selection.
In Some Facts about Centers, Indexicals, and Demonstratives (acl 1991), Passonneau examines how speakers and writers use pronouns like it versus that to manage a reader's focus. The paper introduces the concept of intra-textual deixis, where a pronoun does not point to a pre-existing physical object, but instead creates a brand-new referent out of a preceding clause or action.
- "It" and the Local Center: Choosing the pronoun it often establishes or maintains a Local Center—a temporary, highly focused pointer in the discourse. This allows an abstract concept or a whole event from a previous sentence to be treated as a concrete object of ongoing discussion.
- "That" and Recontextualization: Conversely, choosing that often signals a shift, indicating that the item being referenced is peripheral or that the discourse is moving to a new focus.
By tracking these linguistic cues, readers dynamically update their mental maps, deciding which concepts are central to the current context and which can be sidelined.
Communicative Deixis and Structural Artifacts
The need to map language to non-physical spaces extends beyond fictional story worlds into structural and pedagogical texts. In a textbook or a technical document, phrases like "this section," "the diagram above," or "those figures" do not point to a physical room or a narrative universe; they point to the document itself.
Wilson et al. formalize this phenomenon as Communicative Deixis (CD) in Determiner-Established Deixis to Communicative Artifacts in Pedagogical Text. Here, the text treats its own structural components—paragraphs, tables, chapters—as physical artifacts within a shared space.
Just as a child must learn to shift from physical pointing to narrative world-building, natural language processing (NLP) systems must learn to resolve these communicative references. Standard linguistic tools frequently fail to identify the exact text span or image a phrase points to. To truly understand text, an agent—whether human or artificial—cannot treat sentences as isolated strings. They must treat the document as a structured environment, navigating its internal layout just as one would navigate a physical room.
Key Foundational Work
BruceLanguage Generation: The Linguistics Component (1977) — contributed 14 concept elements to this direction.
From Conversations to Stories: An Analysis of Differences Between Oral and Written Language Comprehension (1977) — contributed 11 concept elements to this direction.
From Conversations to Stories: An Analysis of Differences Between Oral and Written Language Comprehension (1977) — contributed 10 concept elements to this direction.