TL;DR
This paper explores how to formally model "unbounded stress" in languages, a phenomenon where stress can fall far from the beginning or end of a word. The authors propose a new class of functions called "input tier-based input strictly local" (I-TISL) to capture "default-to-opposite-side" stress systems, showing it's more flexible than previous models. They also demonstrate that "default-to-same-side" stress systems are not "subsequential" and argue for a specific definition of "weak determinism" to account for them.
What Problem Does This Paper Address?
Phonology, the study of sound systems in language, often deals with how stress is assigned to syllables within words. "Unbounded stress" refers to stress patterns that aren't simply tied to the beginning or end of a word, making them challenging to model with traditional computational linguistic tools. Previous work has tried to capture these patterns using "subregular" formalisms, which are less powerful than full Turing machines but more expressive than regular languages.
Specifically, this paper focuses on two types of unbounded stress:
- Default-to-Opposite-Side (DO): Stress falls on the syllable closest to one edge (e.g., leftmost) if it meets a certain criterion (like being a long vowel). If no such syllable exists, stress defaults to the opposite edge (e.g., rightmost).
- Default-to-Same-Side (DS): Similar to DO, but if the criterion isn't met, stress defaults to the same edge (e.g., leftmost).
The authors aim to:
- Develop a formal framework within subregular phonology that can accurately model these unbounded stress systems.
- Investigate the expressive power of different subregular function classes and their suitability for these phenomena.
- Compare different definitions of "weak determinism," a concept used to describe functions that are not strictly sequential but still have some predictable behavior.
Methods
The paper introduces and utilizes several key formalisms:
- Subregular Functions: These are string-to-string mappings that are computationally less powerful than general finite-state transducers but more powerful than regular languages. They are often used to model phonological processes that exhibit local dependencies.
- Input Strictly Local (ISL) and Input-Output Strictly Local (IOSL) Functions: These are specific classes of subregular functions that capture local dependencies based on prefixes of the input and/or output strings.
- Tier Projection: This is a mechanism where certain symbols in an input string are "projected" to a special tier, and subsequent processing only considers these projected symbols. This is crucial for modeling phenomena where only specific types of segments (e.g., dominant syllables) are relevant for stress assignment.
- Input Tier-Based Input Strictly Local (I-TISL) Functions: This is a novel class of functions proposed by the authors. It extends the ISL framework by incorporating a flexible tier projection mechanism. This allows the function to consider local context on a projected tier, making it more powerful than standard ISL functions.
- Subsequential Functions: A function is subsequential if it can be computed by a finite-state transducer that emits output symbols deterministically as it scans the input. This is a key property for modeling processes that can be determined by scanning from one direction.
- Weak Determinism: The paper discusses two definitions:
- Heinz and Lai (2013): A function is weakly deterministic if it can be decomposed into two functions, one left-subsequential and one right-subsequential, with constraints on how they interact.
- McCollum et al. (2018): This definition refines the concept by focusing on "non-interaction" between the decomposed functions, ensuring they don't influence each other's behavior in a way that violates the overall deterministic nature of the process.
The authors analyze the stress systems of Abkhaz (DO) and Lhasa Tibetan (DS) by modeling them as string-to-string functions and then determining which subregular class they belong to.
Datasets
The paper does not introduce new datasets. Instead, it uses existing linguistic data from documented languages to analyze their stress systems. The primary examples are:
- Abkhaz: Data from the standard Abzhuy variety, analyzed by Dybo (1977), Spruit (1986), and Trigo (1992), to illustrate a DO stress system.
- Lhasa Tibetan: Data from the Lhasa variety, as described by Dawson (1980) and cited by Gordon (2007) and Odden (1979), to illustrate a DS stress system.
Key Insights and Ideas
- I-TISL Functions for DO Stress: The authors propose I-TISL functions as a more flexible way to model DO stress systems. This class allows for a more general tier projection than previously assumed, which is necessary for capturing the Abkhaz stress system.
- DS Stress is Not Subsequential: A crucial finding is that the Lhasa Tibetan DS stress system requires scanning the entire input to determine stress, making it impossible to model with a strictly left-to-right (or right-to-left) subsequential function. This highlights the need for more powerful formalisms for certain phonological processes.
- McCollum et al.'s Weak Determinism is More Suitable: The paper argues that the definition of weak determinism proposed by McCollum et al. (2018) is better suited for capturing DS stress systems like Lhasa Tibetan than the earlier definition by Heinz and Lai (2013). This is because the McCollum et al. definition more naturally accommodates the "bidirectional" nature of DS stress, where the system might need to "look ahead" or "look back" to determine stress.
- Connection to Metrical Grid Theory: The paper draws parallels between their formalisms (tiers, tier projection) and the established metrical grid theory in phonology, suggesting that their subregular function approach aligns well with existing linguistic analyses of stress.
Key Results
- The Abkhaz DO stress system can be captured by the proposed I-TISL function class.
- The Lhasa Tibetan DS stress system is not subsequential, meaning it cannot be modeled by a simple left-to-right or right-to-left scan.
- The paper provides a formal argument that the Lhasa Tibetan stress function is interaction-free weakly deterministic according to McCollum et al.'s (2018) definition.
- The authors demonstrate that the Abkhaz stress function is not TISL (tier-based input strictly local) under the primitive tier-projection mechanism, requiring a more expressive tier projection.
Limitations and Future Work
- Weak Determinism Definition: While arguing for McCollum et al.'s definition, the paper notes that it's currently unknown if this class of functions is distinct from general finite-state functions. This is a point for future investigation.
- Markup-Free Criterion: The paper suggests that the "markup-free" criterion for weak determinism (Heinz and Lai) might be too restrictive for stress assignment, as stress inherently involves annotating input. This implies that future work might need to relax such constraints or explore alternative decomposition methods.
- Generalization of I-TISL: The paper focuses on a specific instance of I-TISL for Abkhaz. Further research could explore the full expressive power of I-TISL functions and their applicability to other DO stress systems.
- DS Stress Formalism: While showing DS stress is weakly deterministic, the paper doesn't propose a specific subregular class for it beyond weak determinism. Future work could explore if there are more specific subregular classes that can capture DS systems more efficiently.
Summary updated