English Root Stress across Frameworks – Guierre Meets Analogy – ERSaF

The project aims to study different types of data: data from pronunciation dictionaries and oral data from elicitation studies conducted as part of the project. The data are analyzed using inferential statistics (regression models, conditional inference trees) and computational models based on analogy or Naïve Discriminative Learning.

Our first results confirm certain assertions found in the literature such as the role of opaque prefixation and syllable weight, but they nuance these assertions and invite further exploration:
- on opaque prefixation, our initial results show significant differences between prefixes, suggesting that we should assess the importance of individual prefix properties such as their recurrence rather than rigid categories like “opaque” versus “transparent”. Initial explorations in this area are promising.
- on syllable weight, we find effects in our various studies, but these effects are not always consistent with the literature since the effects are found in all positions, suggesting a more generalized role for this variable and probabilistic in nature.

We have begun to model the data using AML, a computational software based on analogy, which has already enabled us to see that the model performs very well in predicting stress in verbs, and sheds light on the determinants of stress (affixes in particular), which emerge from the model without being given to it beforehand.

We still have work to do on the dictionary data for nouns, for which we started from scratch, unlike verbs, which were already pre-sorted and analyzed. Once this data is ready, we will also analyze it with AML.

The big project over the next few months will be a non-word elicitation study targeting the main factors of interest to us. We had originally planned to travel to Manchester to conduct this study, but eventually decided to conduct it online, which will give us access to a wider variety of participants, and in greater numbers.

As we began to do this year, we will continue to communicate our results and submit articles to specialized journals. We are also organizing a conference in Trier in October, which will bring together researchers from different backgrounds to discuss the issue of non-uniformity in morphophonology.

To date, we have presented thirteen papers at conferences and published two articles. Another article has been accepted and should appear shortly. Five articles are at various stages (writing or revision).

This project serves to (a) promote our current understanding of regularities in English word stress, and (b) foster dialogue and collaboration between theoretical frameworks across research traditions, Guierrian theory and analogy-based theories of phonological grammar.
How is stress assigned in English morphologically simplex words? How do we explain, for example, that words like begín, emít and rètrogréss have stress on the final syllable, but words like lísten, vómit and gállivant? Despite the fact that there is a vast literature and that most scholars agree that English stress is not idiosyncratic, existing theoretical accounts are often not based on large corpora and leave large portions of the data unexplained. The most widely known among these are accounts set within the generative tradition of linguistic theorising, which assume that stress assignment is mainly determined by two groups of factors: morphological complexity and syllable structure.
The proposed project aims to shed new light on regularities in English verb stress, and to ultimately develop a better model, by bringing together perspectives from two research traditions.
• Guierre’s work and that of the Guierrian school represents a phonological framework that is widely used in the French-speaking literature, but is underrepresented especially among the English-speaking contemporary literature.
• ‘Analogy’ refers to a mechanism of linguistic generalisation in which formal properties of an unknown form are learned by comparing that form to similar forms in the mental lexicon. Analogy has been considered the basic foundation of grammatical behaviour in the neogrammarian tradition. In the contemporary literature, analogy plays a central role in usage-based models of (phonological) grammar; computational analogical models have since neogrammarian times developed the approach further and have made it empirically testable.
Our central hypothesis is that, even though Guierrian and a analogy-based theories fundamentally differ in their conceptual architectures, learning from each other will be highly beneficial in developing an adequate model of stress assignment.
We will use a computational analogical mechanism to test the Guierrian system of representation on two types of empirical data on non-uniform word stress in English simplex roots, which have so far defied attempts to account for them within the mainstream generative literature. One type will be dictionary data of existing words; the other will be novel words elicted in a reading study. In terms of the Guierrian framework, this will enable us to test basic assumptions about the relevance and nature of the hypothesised units of representations. In terms of an analogical framework, this will help us develop an adequate theory of representation, something that is currently lacking in analogical accounts of generalisation. We will use our findings to test extant architectural claims, Guierrian, analogical, and generative.