TL;DR
This paper introduces the "Match-Extend" algorithm, a novel, parameter-free method for serializing complex, non-linear phonological structures in "Multiprecedence phonology." It offers a more general and typologically constrained approach to converting abstract representations into pronounceable speech, handling phenomena like reduplication without ad-hoc stipulations.
What Problem Does This Paper Address?
Phonological representations, especially for phenomena like reduplication, are often non-linear and can be modeled as graphs. However, to be pronounced, these structures must be converted into linear strings. Traditional approaches to this "serialization" problem have relied on complex, often language-specific parameters or constraints. This paper tackles the challenge of developing a general, deterministic, and typologically constrained serialization algorithm that can handle the diverse structures found in Multiprecedence phonology without requiring language-specific adjustments. The goal is to bridge the gap between abstract linguistic representations and the physical act of speech production in a principled way.
Methods
The core contribution is the Match-Extend algorithm. This algorithm operates on a graph representation of phonological structures, where segments are nodes and precedence relations are edges. The algorithm works by:
- Initialization: The "StemSet" contains the precedence links of the base form, and the "WorkSpace" initially holds the morphologically added links (e.g., from an affix).
- Match: If two strings in the WorkSpace share a common end, they are collapsed into a single string, removing the redundancy. This step prioritizes matches along longer segments.
- Extend: If no match is found, the algorithm simultaneously extends all strings in the WorkSpace to the right and left by following precedence links from the StemSet.
- Iteration: Steps 2 and 3 are repeated until a single string is formed that spans from the start symbol (#) to the end symbol (%).
Crucially, Match-Extend is designed to be parameter-free and generic, meaning it doesn't require specific constraints for different graph types or languages. It relies on the inherent structure of the graph and the interaction between "Match" and "Extend" operations.
Datasets
Not discussed in the paper. The paper focuses on theoretical development and demonstrates the algorithm's application to various attested phonological patterns and examples from existing literature.
Key Insights and Ideas
- Unified Serialization: The Match-Extend algorithm provides a single, elegant mechanism to serialize a wide range of complex phonological structures, including various types of reduplication, without needing language-specific parameters.
- Typological Constraint: By being parameter-free, the algorithm inherently constrains the possible outputs, making certain unattested patterns impossible. This is a significant improvement over systems that allow for excessive parametrization.
- "Closest Attachment" Principle: The algorithm implicitly captures a principle where morphologically added links attach based on proximity, leading to predictable serialization orders, especially in cases of asymmetry.
- Handling Symmetry: While perfectly symmetrical graphs pose an indeterminacy, the paper suggests that hidden asymmetries (e.g., silent segments) or cyclicity in derivations can resolve these cases, rather than requiring a fundamental change to the algorithm.
- Dissimilation as a Simpler Alternative: For apparent counterexamples involving fixed segments, the paper argues that a dissimilation analysis, integrated with the graph structure, can often provide a more parsimonious explanation than complex, ad-hoc serialization rules.
Key Results
The paper demonstrates that the Match-Extend algorithm successfully derives attested serializations for numerous examples of reduplication and other complex phonological patterns from various languages, including:
- Total reduplication (e.g., Indonesian).
- CV-reduplication (e.g., Tohono O'odham).
- Non-edge fixed segmentism (NEFS) in languages like Javanese, Marathi, Bengali, and Kinnauri.
The algorithm correctly handles cases where specific linearization orders are required, even when they might seem counter-intuitive based on simple traversal order (e.g., following a "backlink" before a "front-link"). It also correctly predicts the serialization for cases involving asymmetrical attachments, such as the Nlaka'pamuctsin example.
Limitations and Future Work
- Symmetrical Graphs: The paper acknowledges that perfectly symmetrical graphs can lead to indeterminate derivations. While suggesting potential resolutions (hidden asymmetries, cyclicity), this remains an area for further research.
- Parallel Structures: The current model primarily assumes linear roots with affixes adding topological complexity. If underlyingly parallel structures are allowed in Multiprecedence Phonology, the Match-Extend algorithm may need amendment or enhancement.
- Empirical Research on NEFS: The authors note that examples of Non-Edge Fixed Segmentism (NEFS) are rare in the literature, and further empirical research is needed to test the algorithm's predictions in this area.
- Interaction with Phonology: While the algorithm handles serialization, its interaction with complex phonological rules (like vowel harmony or allophony) that apply after serialization is an area that requires further exploration. The paper suggests that some apparent counterexamples might be better explained by post-serialization phonological processes rather than issues with the serialization algorithm itself.
Summary updated