Introduction
Native American Sign Language (ASL) users frequently encounter specific challenges when writing in English, particularly regarding the omission of noun phrases (NPs). This phenomenon, known as "illegal NP omission," is a form of language transfer where the discourse-oriented grammar of ASL is applied to the sentence-oriented structure of English. This paper presents an automated method for detecting and correcting these omissions by treating them as "invisible pronouns" and resolving them through an extended focus-tracking algorithm.
The Linguistic Challenge: Discourse vs. Sentence Orientation
In linguistics, languages are often categorized by how they manage information flow. ASL, like Chinese, is a discourse-oriented language. In such languages, a speaker may omit a subject or object if the entity was established as the "topic" in previous discourse. Conversely, English is a sentence-oriented language that generally requires explicit noun phrases for grammatical correctness.
When ASL signers apply native discourse rules to English, they produce sentences that appear ungrammatical to native English speakers. For example, a writer might omit the subject "I" or the object "it" if they believe the context makes the reference clear. The objective of this research is to bridge this gap using Computer-Assisted Language Learning (CALL) tools that can identify these missing elements and suggest appropriate corrections.
Methodology: Extending Sidner’s Focusing Algorithm
The proposed system builds upon the work of Sidner (1979, 1983), utilizing a computational method to track the "focus" of a text—essentially what the text is "about" at any given moment. The algorithm maintains three core components:
- Current Focus (CF): The primary entity the sentence is discussing.
- Actor Focus (AF): The agent or "doer" of the action.
- Potential Focus List (PFL): Other entities introduced that may become the focus in subsequent sentences.
The Extended Approach
To adapt this for NP omission, the authors introduced several key modifications to Sidner's original logic:
- Omissions as Anaphora: The system treats a missing NP as a null pronoun. It then searches the focus history to find the most semantically and syntactically appropriate "filler."
- Preference Ranking: Instead of a simple search, the algorithm prioritizes the previous sentence's CF. If the CF is unsuitable, it moves to the PFL and then to "stacked" (historical) foci.
- Handling Complex Sentences: The authors developed specific rules for conjunctions. For instance, in "X because Y" constructions, the focus typically remains with the "effect" clause (X) rather than the "cause" clause (Y).
- Focus Stacking: Unlike the original algorithm, this version "stacks" both the CF and PFL. This allows the system to handle "topic returns," where a writer briefly deviates from a subject before returning to it.
Data and Qualitative Analysis
The researchers analyzed writing samples from deaf students across several institutions, including Gallaudet University, the National Technical Institute for the Deaf, and the Pennsylvania School for the Deaf. These real-world samples provided the basis for identifying error patterns and testing the algorithm's logic.
The paper demonstrates the algorithm's efficacy through a qualitative walkthrough of a student's essay on budgeting. In one instance, the system correctly identified that in the clause "because _ go out to eat," the missing agent was "I" (the Actor Focus of the preceding clause). In another case—"when I need _"—the algorithm correctly identified "money" as the missing NP by tracking it as the Current Focus, even though other nouns like "home" or "parents" were mentioned more recently.
Key Insights
A significant finding of this research is that language transfer errors are predictable and systematic rather than random. Furthermore, the study reveals that the "Recency Rule"—a common heuristic in pronoun resolution that favors the most recently mentioned noun—is often insufficient for resolving null arguments. Instead, the "Current Focus" (the discourse topic) serves as a far more reliable predictor of what the missing NP should be.
Limitations and Future Directions
While the logic-based approach shows promise, several challenges remain for future development:
- Ambiguity: Resolving sentences where multiple non-agent NPs are missing remains difficult.
- Structural Transitions: The system needs refinement on how to handle paragraph transitions—specifically whether to reset the focus stack or continue it from the previous paragraph.
- Quantitative Scaling: The current work focuses on algorithmic design and logic; future research should provide large-scale quantitative performance metrics, such as F1 scores, to evaluate the tool's accuracy across diverse writing styles.
Summary updated