Inductive learning of locality relations in segmental phonology

McMullin, Kevin; Hansson, Gunnar Ólafur

doi:10.5334/labphon.150

Cited by 17 publications

(13 citation statements)

References 63 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…An example of such a system would be strictly beyond-transvocalic dissimilation, where two segments dissimilate only when they are separated by at least a vowel and a consonant; no dissimilation occurs across just a vowel (i.e., transvocalic contexts). McMullin (2016) and McMullin & Hansson (2019) ran a series of artificial grammar learning experiments, finding that strictly beyond-transvocalic patterns of liquid harmony and dissimilation are not reliably acquired in an experimental setting. Even when given unambiguous training data, participants tended to either infer a more typical unbounded pattern or else not infer any pattern.…”

Section: Subsequential Functionsmentioning

confidence: 99%

Long-distance phonological processes as tier-based strictly local functions

Chandlee

Burness

McMullin

2021

Glossa: A Journal of General Linguistics

View full text Add to dashboard Cite

Whether we analyze phonological processes using a system of rules or constraints, the resulting map from underlying representations to surface pronunciations can be characterized as a function. Viewing processes as mathematical objects in this way allows us to study properties of phonology that hold no matter how it is implemented. Work in this vein has found that a majority of phonological processes only consider information within a finite window, placing them in the highly restrictive class of Strictly Local (SL) functions (Chandlee 2014; Chandlee et al. 2014;2015). Long-distance phonological processes, however, lie outside the capabilities of the SL functions since they consider information that can be arbitrarily distant. The more powerful class of subsequential functions has been offered as a potential upper bound on the complexity of long-distance phonology (Heinz and Lai 2013; Luo 2017; Payne 2017), but we will argue that an even tighter bound is possible. Specifically, by incorporating an autosegmental tier (e.g., Goldsmith 1976) into the structure of an SL function, the non-local information crucial for applying long-distance processes can be rendered local. In addition to assessing the typological coverage of these Tier-based Strictly Local (TSL) functions (Burness and McMullin 2019; Hao and Andersson 2019; Hao and Bowers 2019), we show that they fail to generate a number of pathological patterns that can be characterized as subsequential functions. We therefore conclude that the TSL functions are a better hypothesized upper bound on phonological complexity.

show abstract

Section: Subsequential Functionsmentioning

confidence: 99%

Long-distance phonological processes as tier-based strictly local functions

Chandlee

Burness

McMullin

2021

Glossa: A Journal of General Linguistics

View full text Add to dashboard Cite

show abstract

“…Linguistically, these refinements have garnered interest since the morphological and phonological typology correlates with these refinements, favouring the weakest subclasses in the subregular hierarchy. Experimental work also favours this characterization (Finley 2008;Lai 2015;McMullin and Hansson 2019). Our learning algorithms can be applied to model-theoretic treatments of other linguistic representations such as syntactic trees or autosegmental graphs, which opens a useful direction for future research.…”

Section: Regular and Subregular Patternsmentioning

confidence: 66%

Typology emerges from simplicity in representations and learning

Lambert

Rawski

Heinz

2021

JLM

View full text Add to dashboard Cite

We derive well-understood and well-studied subregular classes of formal languages purely from the computational perspective of algorithmic learning problems. We parameterise the learning problem along dimensions of representation and inference strategy. Of special interest are those classes of languages whose learning algorithms are necessarily not prohibitively expensive in space and time, since learners are often exposed to adverse conditions and sparse data. Learned natural language patterns are expected to be most like the patterns in these classes, an expectation supported by previous typological and linguistic research in phonology. A second result is that the learning algorithms presented here are completely agnostic to choice of linguistic representation. In the case of the subregular classes, the results fall out from traditional model-theoretic treatments of words and strings. The same learning algorithms, however, can be applied to model-theoretic treatments of other linguistic representations such as syntactic trees or autosegmental graphs, which opens a useful direction for future research.

show abstract

“…It is well-known that learning non-adjacent dependencies poses a particular challenge for language learning (Bonatti et al, 2005;Endress & Bonatti, 2007;Frost & Monaghan, 2016;Gomez, 2002;Gomez & Maye, 2005;Misyak et al, 2009;Misyak & Christiansen, 2007;Newport & Aslin, 2004). This is an issue both in terms of non-adjacent dependencies among words in sentences (Wilson et al, 2020), but also within words in phonological patterns like vowel and consonant harmony (Finley, 2011(Finley, , 2015McMullin & Hansson, 2014, 2019, morphological patterns like non-concatenative morphology (Drake, 2018;Fullwood, 2018;Fullwood & O'Donnell, 2013).…”

Section: Segmentation Of Root and Pattern Morphologymentioning

confidence: 99%

Segmentation of root and pattern morphology

Finley¹,

Newport²

2021

Preprint

View full text Add to dashboard Cite

Morphology is the study of how form and meaning are combined to form complex words. While previous studies of morphology learning rely on semantic associations of continuous affixes (e.g., prefixes and suffixes), the present study focuses on the learnability of non-continuous (non-concatenative) forms, without the use of semantic information. We performed three artificial grammar learning experiments testing the types of information that adult, English speaking learners can extract from hearing words made up of CCC ‘roots’. In Experiment 1, participants were exposed to 24 CVCVC words made up of 12 CCC roots and four VV residues, repeated 15 times. In Experiments 2 and 3 the number of CCC items was increased to 72 (repeated five times), but with four additional templates (e.g., CVCCV in addition to CVCVC) in Experiment 2, and the addition of a prefix in Experiment 3. The results were parallel across all three experiments: participants could readily identify familiar items compared to both ungrammatical and novel grammatical items, and could correctly identify novel words compared to ungrammatical items, but only when the ungrammatical item was sufficiently different from the items heard in training. These results suggest that while learners can extract discontinuous information from lexical items, learners rely heavily on their memory for these lexical items, suggesting a possible bias against learning non-concatenative morphology.

show abstract

Inductive learning of locality relations in segmental phonology

Cited by 17 publications

References 63 publications

Long-distance phonological processes as tier-based strictly local functions

Long-distance phonological processes as tier-based strictly local functions

Typology emerges from simplicity in representations and learning

Segmentation of root and pattern morphology

Contact Info

Product

Resources

About