Workshop on The Role of Representation in Computational Phonology

Handouts and slides: https://drive.google.com/drive/folders/1gdHRhoehIBKAnqPygFMIUVZ1DfDycSYj?usp=sharing

We are pleased to announce the OCP23 Satellite Workshop on The Role of Representation in Computational Phonology to be held on 13 January at King's College (University of Cambridge). The workshop is free of charge, but we cannot accommodate more than 40 people. We therefore ask you to contact Yury Makarov for confirmation of your place beforehand.

Phonological representations minimally assume linearly ordered elements, and have been augmented to include hierarchical structure like prosodic categories and non-linear structure like autosegmental tiers. McCarthy’s (1988:84) often quoted adage, “if the representations are right, then the rules will follow” implies the analytic advantage of representations other than strings. This workshop explores their computational consequences.

To give an illustrative example, Slovenian bans [+anterior] stridents before [−anterior] stridents, Slovenian bans [+anterior] stridents before [−anterior] stridents, e.g., [sl-ux] ‘hearing’ [ʃl-iʃ-i] *[sl-iʃ-i] ‘hears’ (Jurgec, 2011:330). This can be modeled by projecting a tier of stridents and banning adjacent segments that disagree in the feature [anterior], e.g., [ʃl-iʃ-i] → [ʃʃ], *[sl-iʃ-i] → *[sʃ]. This approach straightforwardly captures blocking effects by intervening coronal stops, e.g., [sit] ‘full’ [na-sit-iʃ] ‘(you) feed’ (Jurgec, 2011:331), by projecting them onto the tier: [na-sit-iʃ] → [stʃ]. Unsurprisingly, strident affricates participate in harmony, e.g., [ʦepəʦ] ‘foot’ [ʧepʧək] ‘fool-DIM’ (Jurgec, 2011:330). However, if affricates were interpreted as clusters, it would not be possible to model these alternations by banning strings on a projected tier, illustrating an interaction between phonological representations and computational complexity (see Lamont, 2023 for more examples along these lines). Previous work has shown that adopting autosegmental representations reduces the complexity of tonotactics (Jardine, 2016) and that adopting feet reduces the complexity of stress (Koser, 2022), providing computational support for these representations.

We invite abstracts for workshop talks on the role representations from traditional phonological theory play in computational phonology and how computational approaches can inform traditional views of representation. Topics may include, but are not limited to:

  • Phenomena whose analyses are made simpler/more complex by adopting different representations
  • The role of learnability (e.g., Gouskova and Stanton, 2021; Lee et al., 2023) and what structures should/must be encoded in UG
  • Equivalences between different representations (e.g., Strother-Garcia, 2019; Jardine et al., 2021)

We especially encourage submissions that discuss novel data sets and/or representations that are not widely assumed in the phonological literature. Submissions do not need to be couched in model theory/formal language theory; we invite perspectives from various theoretical backgrounds.

References

Gouskova, Maria and Juliet Stanton (2021). Learning complex segments. Language 97(1). 151–193.

Jardine, Adam (2016). Locality and non-linear representations in tonal phonology. Ph.D. dissertation, University of Delaware.

Jardine, Adam, Nick Danis and Luca Iacoponi (2021). A formal investigation of Q-Theory in comparison to Autosegmental Representations. Linguistic Inquiry 52(2). 333–358.

Jurgec, Peter (2011). Feature spreading 2.0: A unified theory of assimilation. Ph.D. dissertation, University of Tromsø.

Koser, Nate (2022). The computational nature of stress assignment. Ph.D. dissertation, Rutgers, The State University of New Jersey.

Lamont, Andrew (2023). Phonotactics conspire to reduce computational complexity. Unpublished manuscript, University College London. Available at https://aphonologist.github.io/papers.

Lee, Seung Suk, Alessa Farinella, Cerys Hughes and Joe Pater (2023). Learning stress with feet and grids. In Noah Elkins, Bruce Hayes, Jinyoung Jo and Jian-Leat Siah (eds.), Proceedings of the 2022 Annual Meeting on Phonology. Washington, D.C.: Linguistic Society of America. https://doi.org/10.3765/amp.v10i0.5441.

McCarthy, John J. (1988). Feature geometry and dependency: A review. Phonetica 45. 84–108.

Strother-Garcia, Kristina (2019). Using model theory in phonology: A novel characterization of syllable structure and syllabification. Ph.D. dissertation, University of Delaware.

Key dates

1 July 2025: submissions open

1 August 2025: submissions closed

September 2025: accepted submissions' authors notified

13 January 2026: workshop takes place in Cambridge

14–16 January 2026: main conference takes place in Cambridge

Programme

9:00-9:45Coffee
9:45-10:00Formants are better predictors of vowel markedness than features
Benjamin Storme
10:00-10:30Phonetic substance is encoded in the phonological featural system: neural network modeling of vowel harmony and vowel disharmony
Shuang Zheng and Youngah Do
10:30-11:00Discrete Interpretations of Continuous Representations in Model-Theoretic Computational Phonology
Scott Nelson
11:00-12:00Invited talk: Representing phonology in the formal comparison of phonological representations
Nick Danis
Recent work in the application of finite model theory to phonological theory provide, among other things, a new framework for the meta-theoretical comparison of different proposed phonological representations (Strother-Garcia 2019; Oakden 2020; Jardine et al. 2021; a.o.). Within this framework, a logical transduction, or translation, between two theories (defined as finite models) indicate (some level of) notational equivalence. The present work expands this framework to include mechanisms of comparison beyond the transduction that more transparently incorporates linguistically relevant properties such as phonological contrast, natural class behavior, and featural constituent behavior (as a few examples). Additionally, while the power of the logical language (e.g. quantifier-free vs. full first order logic) used to define the rules of some transduction has often been used as a sufficient indicator of notational equivalence (such that using a less powerful logic implies less substantive differences between two representations), two representations can differ in these linguistic properties while still being bi-interpretable under weak logic. Ultimately, the syntax of the rules of the transduction offer a more linguistically substantive indication of equivalence than the general power of the logic itself. The comparison of these relevant linguistic properties are cast in a framework similar to Miller (2001)'s model-theoretic approach to comparing different syntactic theories. This additional mechanism augments the insights gained from the transduction alone in a way that better captures the linguistic intent of the representations themselves, without abandoning the formality of a model-theoretic approach.
12:00-1:00Lunch
1:00-1:30The autosegmental route to anbn patterns
Daniel Gleim
1:30-2:00Kalabari tonotactics require forbidden substrings not substructures
Nicholas Rolle and Adam Jardine
2:00-2:30Comparing Tonotactic Learning over Strings and Autosegmental Representations
Han Li and Jeffrey Heinz
2:30-3:15Coffee
3:15-3:45Learning and Representing Process Interaction Through Simplex ISL Transducers
Chenli Wang and Adam Jardine
3:45-4:15Assimilation, deletion, complementarity, and licensing
Eric Baković
4:15-4:30Where purely representational approaches fail: construction-specific phonology
Bartlomiej Czaplicki
4:30-4:45Alignment Theory in Optimality Theory: A Proposal
Mathilde Hutin
4:45-6:00Drinks reception