A canonically agglutinative language or morphological pattern is traditionally analyzed as building words out of independent morphemes. Using data from Choguita Rarámuri (Uto-Aztecan), we attempt to quantify this notion by examining the extent to which meanings are predictable from their exponents without reference to context. We show that two-layer connectionist networks, computational models that map form onto meaning directly, can be used for this purpose. We also show that learning the meanings of morphemes can pose significant challenges to such models and constrains the design of the learning algorithm. In particular, models trained to equilibrium tend to focus on unreliable cues to the meanings they try to predict, especially when trained on a small corpus typical of underresourced languages. Some of these issues can be alleviated by a slow learning rate. However, one issue — which we call the problem of spurious excitement — is shown to be inherent to the learning algorithm, and always arises by the time the model achieves equilibrium. Spurious excitement means that a cue becomes associated with a meaning that it does not co-occur with, simply because of co-occurring with cues that disfavor the meaning. This case raises larger implications with respect to the type of learning mechanism involved in the acquisition of natural languages. Solutions to spurious excitement are discussed. The logistic activation function is shown to improve the performance of the model in detecting reliable cues to meanings that recur across many word types (i.e., cues of high type frequency), as well as eliminating spurious excitement.

We develop a large set of pseudowords that systematically varies length and phonotactic probability and obtain acceptability ratings using an online interface. We find that phonotactic likelihood and the presence of an apparent morphological parse both significantly predict acceptability; pseudowords containing known morphemes are more acceptable than otherwise comparable pseudowords that do not. We find support for the conjecture that novel words with apparent morphology are advantaged as additions to the lexicon. The resulting lexicon, as observed, is one in which long words are not a random sampling of phonotactically acceptable wordforms, but instead tend to be completely or partially decomposable into morphemes.

The past few years have led to the widespread recognition that morphology is an independent domain of language functioning in dynamic interdependence with more familiar domains such as phonology and syntax. This has permitted nuanced research into the organization of morphological systems as well as the development of hypotheses concerning factors responsible for such organization. In this chapter we compare two classes of hypotheses — adaptive explanations and neutral ones — for attested differences in morphological complexity claimed to correspond with sociocultural and demographic factors. While both examine language change as a (cultural) evolutionary process, we argue that much recent work on adaptive hypotheses for morphological complexity has been uncritically adaptationist, neglecting key results and lessons from population genetics about how to study evolutionary systems. Finally, we argue that neutral explanations are presently more likely explanations for the apparent association of morphological complexity and smaller, historically more isolated populations and should a priori be preferred over adaptive explanations unless and until a high evidential burden has been met.

Competing models of lexical access propose contrasting roles for morphological structure in word recognition. Whole-word models suggest that there are no separate representations for morphemes (e.g., Tyler et al. ); decomposition models posit that words are recognized by accessing their constituent morphemes (e.g., Taft et al. ); and hybrid models incorporate both pathways to recognition (e.g., Bertram et al. ). The relative productivity of a word’s derivational affixes may also play a role: words with unproductive affixes are processed holistically whereas words with productive derivational affixes are processed as a function of their morphemes (e.g., Balling and Baayen ). In this paper, we examine the role of the Semitic consonantal root, known to be a route for lexical retrieval, and its interaction with relative binyan productivity. Extending the methodology developed by Wray () for Jordanian Arabic, we investigate the Semitic language Maltese. Based on two auditory lexical decision experiments, we find a reverse base frequency effect in a productive binyan (words with more frequent roots are recognized more slowly than words with less frequent roots), and in two less productive binyanim we find no base frequency effect. This supports the validity of models in which morphological decomposition is relevant strictly for productive affixes.

A number of scholars have argued for the need to postulate principles of rule combination in morphological theory; according to such principles, two rules may combine to produce a more complex rule. Several kinds of evidence motivate the postulation of such principles, which afford new and revelatory explanations for a range of familiar morphological phenomena. Central to these explanations is a set of four characteristics (component independence, phonological transparency, semantic transparency, and domain subsectiveness) that combined rules possess by default but from which they may also deviate. This set of characteristics has both synchronic and diachronic significance. Synchronically, they elucidate the nature of potentiation, the relation between two affixes A and B such that stems created by means of A extend the domain of stems to which B may subsequently attach (Aronoff , Williams ). Diachronically, they illuminate the nature of affix telescoping, the diachronic correspondence of a sequence of two affixes at one stage in a language’s history to a single affix at a later stage (Booij , Haspelmath ). The evidence discussed here lends additional strength to the conclusion that principles of rule combination are a necessary addition to morphological theory.

This paper examines the emergence of a pattern that Stump and Finkel () dub Marginal Detraction: a tendency in inflection class systems for low type frequency (i.e., irregular) classes to disproportionately detract from the predictability of regular classes. We ask: What factors lead to the emergence (and sometimes non-emergence) of Marginal Detraction? We use an iterated agent-based Bayesian learning model to simulate the conditions for analogical restructuring of inflection classes over time. Input to the model consists of artificial inflection class systems that vary in how the classes overlap — their network structure. We find that network properties predict whether the Marginal Detraction distribution emerges within the model. We conclude that languagespecific network properties shape local interactions among words and thereby likely play a significant role in analogical inflection class restructuring and the emergence (or non-emergence) of global properties of inflectional systems.