Historical and psycholinguistic perspectives on morphological productivity: A sketch of an integrative approach

TL;DR

This paper investigates morphological productivity from cognitive‑computational and diachronic angles using the Discriminative Lexicon Model (DLM) and a Thomas Mann case study. It shows that productive generalization hinges on systematic form–meaning mappings, with Finnish inflection, Malay derivation, and English compounding providing cross‑linguistic contrasts in learnability and generalization. Across languages, centroids of case–number or prefix semantics anchor the contribution of sublexical n‑grams, enabling varying degrees of generalization; Mann’s writings reveal a striking gap between abundant reading input and scarce production of novel derivations, though his production correlates with input novelty. The work highlights two types of productivity—systemic generalization and bricolage—and argues for agent‑based modeling to capture dynamic, socially embedded changes in lexical innovation.

Abstract

In this study, we approach morphological productivity from two perspectives: a cognitive-computational perspective, and a diachronic perspective zooming in on an actual speaker, Thomas Mann. For developing the first perspective, we make use of a cognitive computational model of the mental lexicon, the discriminative lexicon model. For computational mappings between form and meaning to be productive, in the sense that novel, previously unencountered words, can be understood and produced, there must be systematicities between the form space and the semantic space. If the relation between form and meaning would be truly arbitrary, a model could memorize form and meaning pairings, but there is no way in which the model would be able to generalize to novel test data. For Finnish nominal inflection, Malay derivation, and English compounding, we explore, using the Discriminative Lexicon Model as a computational tool, to trace differences in the degree to which inflectional and word formation patterns are productive. We show that the DLM tends to associate affix-like sublexical units with the centroids of the embeddings of the words with a given affix. For developing the second perspective, we study how the intake and output of one prolific writer, Thomas Mann, changes over time. We show by means of an examination of what Thomas Mann is likely to have read, and what he wrote, that the rate at which Mann produces novel derived words is extremely low. There are far more novel words in his input than in his output. We show that Thomas Mann is less likely to produce a novel derived word with a given suffix the greater the average distance is of the embeddings of all derived words to the corresponding centroid, and discuss the challenges of using speaker-specific embeddings for low-frequency and novel words.

Figures (10)

• Figure 1: By-class accuracy of a frequency-informed comprehension model on held-out data as a function of three productivity measures. Dots represent inflectional classes.
• Figure 2: Position of 34,262 Finnish nominal forms in a 2-dimensional t-SNE plane. Clustering is by case. The cluster for essive singular is presented in red, and the cluster for illative plural is given in blue. The black points represent the centroids of these clusters.
• Figure 3: Proportion of trigrams in a word for which the correlation with the compound embedding exceeds the correlation with the embedding of a left (LC) or right (RC) constituent, broken down by boundary trigrams (boundary tri), trigrams preceding the boundary (left tri) and trigrams following the boundary (right tri).
• Figure 4: Correlation heatmap of prefix centroid embeddings and a subset of the row vectors of the F matrix that maps the form vectors (4-gram) to its meaning (FastText embedding) in the DLM. The set of 4-grams presented correspond to at least one of the prefixes. Darker hues, compared to lighter hues, indicate a stronger correlation between the embeddings of the centroid and row vectors of the F matrix. The strongest correlations are present for the 4-grams that overlap most with the prefix, indicating that it is the prefixal 4-grams that contribute most to realizing the meaning of the centroid.
• Figure 5: Partial effects of the GAMM fitted to the training data. Left column: Partial effect of word frequency on RT (top). Partial effect of word length on RT (bottom). Rugged lines on the x-axes represent the data. Right column: Partial effect of the interaction between Target Correlation$_{\text{train}}$ of the DLM and the correlation between each derived word and their centroid. Red hues denote shorter RTs, orange/yellow hues indicate longer RTs. Black dots represent the data.