Logical Transductions for the Typology of Ditransitive Prosody

Mai Ha Vu, Aniello De Santo, Hossep DolatianNAACL 2022

TL;DR

This paper tackles the challenge of formally mapping syntactic structures to prosodic structures in ditransitive sentences. Using logical tree transductions, the authors provide a computational framework to model the diverse ways languages group words into prosodic phrases, revealing how subtle syntactic details are crucial for accurate prosody.

What Problem Does This Paper Address?

In linguistics, understanding how the spoken rhythm and intonation of a sentence (prosody) relate to its grammatical structure (syntax) is a fundamental question. While it's clear that prosody isn't a direct mirror of syntax, the precise mechanisms of this mapping are complex and vary across languages. This paper focuses on ditransitive sentences (e.g., "She gave Mary books"), which are known to exhibit significant cross-linguistic variation in their prosodic phrasing. The authors aim to provide a formal, computational model that can capture these variations, building on existing typological work and highlighting the theoretical assumptions needed for such a model. They specifically address the gap in computational machinery for sentence-level phonology, which has historically lagged behind word-level phonology.

Methods

The core methodology employed is logical tree transductions. This approach uses Monadic Second-Order (MSO) logic to define transformations between tree structures. The authors leverage this to model the mapping from syntactic trees to prosodic phrase trees.

  • Standard: The use of logical transductions for phonological processes is established, particularly for word-level phenomena.
  • Novel: The paper extends this to sentence-level prosody, specifically focusing on the complex syntax-prosody interface in ditransitive constructions. They define a set of predicates that capture key syntactic features relevant to prosody, such as overt material, headedness, tree geometry, argument structure, head movement, and linearity. These predicates are then used to construct logical transductions for different prosodic typologies.

Datasets

Not discussed in the paper. The paper relies on existing typological data and examples from the literature, particularly from Kalivoda (2018), rather than introducing new datasets or benchmarks.

Key Insights and Ideas

  • Formalizing Mismatches: Logical transductions provide a powerful tool to formally capture the "mismatches" between syntactic and prosodic constituents, which are common in languages.
  • Syntactic Details Matter: The paper demonstrates that seemingly minor details of syntactic structure, such as head movement and the precise geometric relationships between nodes, are crucial for determining prosodic phrasing.
  • Category Blindness Hypothesis: The authors explore the hypothesis that prosody is "blind" to syntactic category labels (like NP vs. vP), relying instead on structural properties. However, they note that the data they analyze might be biased towards simpler structures, and more complex nominals could challenge this hypothesis.
  • Need for Explicit Assumptions: The formalization process forces researchers to make explicit theoretical assumptions about syntax and prosody that are often left underspecified in linguistic literature.

Key Results

The paper successfully defines logical transductions that can model four distinct prosodic typologies for ditransitive constructions observed in SVO languages (Ewe, Chimwiini, Kimatuumbi, Zulu). These transductions demonstrate how different syntactic features are utilized to derive the observed prosodic groupings. For instance, they show how the "separated" prosody in Ewe ((V)(N)(N)) can be derived by treating each NP as a separate prosodic phrase, while the "all-merged" prosody in Zulu ((VNN)) requires a single prosodic phrase derived from the entire verb phrase. The paper also establishes commonalities, such as the mapping of the CP node to an intonational phrase (iP) and overt terminal items to prosodic words (PW).

Limitations and Future Work

  • Head Movement Locality: The assumption that head movement is always local (bounded within the vP) is a simplification. While justified by the data, future work could explore how unbounded head movement might be handled.
  • Recursive Structures: The formalization of "recursive" prosody (e.g., in Kimatuumbi) is presented as a starting point, and extending it to deeper levels of embedding remains an open question.
  • Category Blindness: The authors acknowledge that their analysis of category blindness might be limited by the type of data available (bare NPs). Future research should investigate more complex nominal structures to test this hypothesis more rigorously.
  • One-to-Many Mappings: The paper notes that while they avoid one-to-many mappings (one syntactic XP mapping to multiple prosodic phrases), this is a common theoretical preference and might need re-evaluation.
  • Future Directions: The authors suggest that their work provides a baseline for extensive formalization of syntax-prosody mismatches, opening avenues for computationally informed empirical predictions and further theoretical refinement. They also point to the potential for integrating their logical transductions with existing work on tree transducers and quantifier-free string transductions.

Summary updated

BibTeX
@inproceedings{vu2022logical,
  author = {Mai Ha Vu and Aniello De Santo and Hossep Dolatian},
  title = {{Logical Transductions for the Typology of Ditransitive Prosody}},
  booktitle = {Conference of the North American Chapter of the Association for Computational Linguistics},
  year = {2022},
}

Loading related content...

Figure