The natural stability of autonomous morphology

Abstract

Autonomous morphology, such as inflection class systems and paradigmatic distribution patterns, is widespread and diachronically resilient in natural language. Why this should be so has remained unclear given that autonomous morphology imposes learning costs, offers no clear benefit relative to its absence and could easily be removed by the analogical forces which are constantly reshaping it. Here we propose an explanation for the resilience of autonomous morphology, in terms of a diachronic dynamic of attraction and repulsion between morphomic categories, which emerges spontaneously from a simple paradigm cell filling process. Employing computational evolutionary models, our key innovation is to bring to light the role of `dissociative evidence', i.e., evidence for inflectional distinctiveness which a rational reasoner will have access to during analogical inference. Dissociative evidence creates a repulsion dynamic which prevents morphomic classes from collapsing together entirely, i.e., undergoing complete levelling. As we probe alternative models, we reveal the limits of conditional entropy as a measure for predictability in systems that are undergoing change. Finally, we demonstrate that autonomous morphology, far from being `unnatural' (e.g. \citealt{Aronoff1994}), is rather the natural (emergent) consequence of a natural (rational) process of inference applied to inflectional systems.

Figures (30)

• Figure 1: Paradigm cell filling mechanism. Rows are lexemes, columns are cells; the focal cell of the focal lexeme, marked '?', is to be filled. (a) Select a pivot cell and examine its contents in all lexemes. (b) Evidence lexemes are those whose pivot cell matches that of the focal lexeme. (c) Examine the focal cell contents of the evidence lexemes. (d) Exponents score +1 for each token; select the highest-scoring exponent (in this case, x).
• Figure 2: Replication of AckermanMalouf2015's (AckermanMalouf2015) model, with tidying-up (i.e., deletion of duplicate lexemes after each change). Evolution of 100 initial lexemes with 8 cells and 5 exponents available in each cell. Black lines show mean values of 100 simulation runs. Grey ribbons indicate 90% of runs' variation. All simulations ran for 2,500 cycles, indicated on horizontal axes. Note most vertical axes are non-linear, to enhance the visibility of model behaviour at low values.
• Figure 3: Replication of AckermanMalouf2015's (AckermanMalouf2015) model, without tidying-up (i.e., identically-inflected lexemes are tolerated, not deleted). 100 simulations for 10,000 cycles.
• Figure 4: Eight snapshots evenly spaced between cycle 0 (leftmost) and cycle 10,000 (rightmost) from one simulation of AckermanMalouf2015's (AckermanMalouf2015) model without tidying-up. Each snapshot shows 100 lexemes in rows, 8 cells in columns. Distinct exponents in each cell are indicated by shading.
• Figure 5: Diagrammatic representation of (a) entropy (circles) and conditional entropy (in grey), and two scenarios in which conditional entropy declines: (b) due to increased overlap, (c) due to decreased overall entropy.