Research
A pervasive question within not only linguistics, but cognitive science in general, is the nature of the mental structures that are formed during learning, used to categorize and label objects in the real world, and that act as the symbols manipulated by mental processes. In phonology this question reaches all the way through speech recognition, the selection of articulatory targets during speech production, the way languages differ in carving up the acoustic space into contrastive sounds, and how those divisions can change over time: categories merging, splitting, or shifting. My research focuses on the interface between phonetics and phonology, and the means by which phonetic representations become phonological representations. This statement presupposes the fact that a distinction can be made between phonetics and phonology, and that linguistic phenomena can be sorted into one domain or the other with some degree of confidence. Because neither of these things is necessarily true, the pursuit of the question of representational change must begin with several prior questions, such as: how phonetic and phonological representations differ, if they do, and how to distribute the explanatory burden between representations, and operations on those representations.
TYPOLOGY & CORPORA
LABORATORY EXPERIMENTS
SIMULATIONS & THOUGHT EXPERIMENTS
Typology & Corpora
Claims about representational structure are inextricably linked to the rest of the theoretical commitments one adopts. In linguistics, more so than many other fields, the evidence used to test a given theory is itself a product of that theory. This means that 'empirical facts' cannot be taken at face value. Re-analysis of original source materials yields alternative proposals about what languages really look like at both the phonological and phonetic levels.
- Epenthesis:
There is an inherent analytic ambiguity involved in deciding whether a given alternation pattern should be analyzed as epenthesis (in inter-vocalic context), or deletion (in pre-consonantal context). Different criteria alter -- sometimes drastically -- what we take to be the typology of consonant epenthesis, and can falsify claims about universal markedness hierarchies for epenthetic segments. Re-analyzing the data from 53 languages that have been claimed to demonstrate epenthesis, I have found no clear preference for coronal segments. In fact, many cases provide poor, or only ambiguous, support for an intervocalic epenthesis analysis at all, suggesting that epenthesis of consonantal segments may be much rarer than previously thought.
Morley, R. Deletion or Epenthesis? On the Falsifiability of Phonological Universals. Lingua 154 (2015): 1-26. [Appendix]
- Vowel Lengthening:
It is a well-known fact that vowels before word-final voiced obstruents are longer than vowels before word-final voiceless obstruents, and that this effect is particularly strong in English. However, the Buckeye Corpus of Conversational American Speech shows no consistent duration difference between these two contexts in CVC words, even after controlling for word frequency, consonant place of articulation, vowel quality, speaking rate, following context and other factors. Instead, the data support the interpretation that the vowel lengthening effect is an artefact of laboratory speech, where vowel durations can be several times longer than what is found in the majority of the corpus. Vowel duration affects the perception of the following consonant duration, and both these measures affect the perception of "voicing" on obstruents in English.
Morley, R.L.,Smith, B., Kohnlein, B., and Kim, N. From Consonant Voicing to Vowel Length: Invariant and co-varying cues to final "voicing" in American English. Poster presented at the Conference on Laboratory Phonology. Lisbon, Portugal.(2018)
Laboratory Experiments
Vowel Duration:
Production experiments in which speaking rate is systematically varied show how word-final consonants and their preceding vowels lengthen differentially in CVC words ending in voiceless obstruents compared to those ending in voiced obstruents. Vowel duration is longer preceding voiced obstruents, as predicted. However, the difference in vowel duration between the voiced and voiceless pairs is quite small at fast speaking rates, and increases with decreasing speed. This trajectory is mirrored by the behavior of the voiced obstruents, which lengthen less and less as speaking rate decreases. Voiceless obstruents, on the other hand, can be seen to lengthen at a fairly constant rate, and to reach durations comparable to those of some vowels. These data support the hypothesis that voiced obstruents are constrained from lengthening past a certain point, and that vowels in the same syllable expand more to compensate. This phenomenon explains the apparent vowel lengthening effect as due, not to the actual voicing difference between the obstruents, or to any acoustic or articulatory universals, but to the interaction of prosodic lengthening with inherent segmental specifications of duration.
Artificial Grammar Learning Experiments: participants are trained on auditory word-picture pairs in a "new language" and told that they will be asked questions about the language later. The training stimuli are designed to conform to a grammatical rule chosen by the experimenter, but which is not explicitly taught to the participant. Testing procedures may be perception-based, e.g., two-alternative forced choice tasks ("chose the word that you think the speaker might have said"), or production-based ("produce the plural of the word you just heard").
- Nasalization:
Learners can encode novel phonetic cues (degree of vowel nasalization) which are differentiable only by quite small acoustic differences, even when those differences are not used contrastively either in their native language or the artificial language of the experiment. They are also able to consistently associate those cues with more abstract linguistic structure, namely, the presence or absence of a morphological juncture.
Morley, R.Rapid learning of morphologically conditioned phonetics: vowel nasalization across a boundary. In A. Yu (ed.), Origins of Sound Change: Approaches to Phonologization. Oxford: Oxford University Press (2013) , pages 181-198.
- Epenthesis:
Phonetically natural tokens -- vowels separated by homorganic glides -- are overwhelmingly treated by listeners as the outcome of phonetically predictable coarticulation over vowel-vowel sequences. This bias can be completely reversed, however, by including an unambiguously phonemic glide in training, but only if the phonemic analysis allows the learner to assume a single allomorph. Phonological learning on the other hand is much more idiosyncratic, if it can be said to occur at all. There is no evidence that learners in any of the conditions acquire a rule of epenthesis, suggesting, at least, that re-analysis, or "rule inversion", requires a number of incremental changes in the distributions of allomorphic variants. However, participants generally show a bias to generalize, if not uniformity in how they do it.
Morley, R.L. From Phonetics to Phonology: Learning Epenthesis. In Proceedings of the 47th Meeting of the Chicago Linguistics Society. (2011) 47(2),pp.137-151.
Morley, R.L. Is Phonological consonant epenthesis possible? A series of artificial grammar learning experiments Phonology 35 (2018): 649-688.
Simulations & Thought Experiments
Computational implementation of linguistic theories often reveals important elements that have been left unspecified, as well as covert assumptions that may be based on the author's intuitions. There is a large amount that remains unknown about language learning, mental representations, and shifts in sound categories. Simulations allow us to explore this large parameter space productively.
- Grammar Change: It has been claimed that synchronic phonological grammars are more regular than would be expected under natural circumstances of change and evolution. However, I have found that the assumptions underlying this claim cannot be made consistent with minimally plausible models of learning, lexical inventories, phonetics, and sound change itself.
- Sonority-To-Stress:
Learning a stress rule from contradictory data
Morley, R. Can Phonological Universals be Emergent? Modeling the space of sound change, lexical distribution, and hypothesis selection.
Language 91:2 (2015): e40-e70 Separate Appendix of all derivations in text. Language 91:2 (2015): s1-s20
- Palatalization:
Simulations of phoneme split as re-categorization of phonetic tokens
Morley, R.L. Implications of an Exemplar-Theoretic Model of Phoneme Genesis: A Velar Palatalization Case Study. Language and Speech (2014) 57:1, pages 3-41
- Epenthesis:
Necessary and sufficient conditions for synchronic epenthesis to result from "rule inversion" of diachronic deletion
Morley, R.L. The Emergence of Epenthesis: An Incremental Model of Grammar Change. Language Dynamics and Change 2 (2012), pages 59-97.
- Nasal Vowels:
The end of this work models a change in which allophonically nasalized vowels become contrastive. This model achieves stability, multiple possible end states, and theoretical consistency by discarding the assumption that there is a single representational structure both before and after change.
Morley, R.L. Sound Structure and Sound Change: A Modeling Approach. Conceptual Foundations of Language Science Monograph Series. Language Science Press (2019)
- Representations: Language is a complex system, tied to perception, learning, memory, and social organization. Because of this, at a certain point, the standard modular approach becomes an obstacle to understanding language at a deeper level. In comparing findings across multiple sub-domains of linguistics the disparity in theoretical assumptions is striking. Developing a single framework that can accommodate the large body of knowledge currently available to us is likely to be the only way to finally make progress on many problems in linguistics that have remained intractable.
Morley, R.L. Sound Structure and Sound Change: A Modeling Approach. Conceptual Foundations of Language Science Monograph Series. Language Science Press (2019)
|