The Mental Lexicon - Volume 19, Issue 2, 2024

Through a consideration of the noun inflections of Finnish, we examine Butterworth’s (1983) full-listing hypothesis (FLH), that whole-word forms are stored in the mental lexicons of language learners/users. While previous studies have consistently confirmed that complex derivational constructions are stored/retrieved this way in Finnish, the current consensus is that the inflectional forms are not, primarily because of the extremely large number that Finnish grammar projects. We challenge this reasoning in favor of a new theory that posits a dual lexicon for inflections, one for the full forms of stems and a parallel one for inflectional affix strings. To test this, we performed a segmentation experiment assessing the relative salience of morpheme and syllable boundaries in Finnish nouns, and also examined a large corpus of quasi-natural spoken dialogues for further evidence. Together, our findings demonstrated that the salience of the boundaries we examined was consistent with our expectations, that the number of inflected forms used in our corpus was only a tiny fraction of those predicted by rule, and that their productivity was highly correlated with their frequencies. While this evidence does not fully confirm the dual lexicon hypothesis, it does strongly indicate that no online grammatical rules are involved.

Many English words contain historical roots that do not occur as free morphemes (e.g., nov in innovate, dict in verdict). These words often retain an appearance of compositionality and are associated with effects on lexical processing (Pastizzo & Feldman, 2004; Taft & Forster, 1975), but frequently their roots are difficult to identify without recourse to historical etymologies, and they are semantically opaque and unproductive. More practically, although such words are prominent in academic vocabulary, they are often difficult to learn, and instruction inspired by their apparent morphological structure has yielded mixed results (McKeown et al., 2018). We explore these psycholinguistic and educational challenges through a dynamic view of the mental lexicon (Libben, 2022), understanding morphological resources as gradient, emergent, and contextually adaptable for meaning making. We quantified bound roots’ morphological families by training an unsupervised parser on a lexicon approximating that of an educated English user, and then assessing polysemy and coherence of roots’ meanings, using vector semantic representations. Testing against behavioral data supported the validity of these measures, suggesting new ways of measuring the properties of bound roots independent from etymological data and demonstrating sensitivity even to unproductive morphological structure, that can support academic vocabulary development and meaning-making.

Morphological priming and paradigm size effects have been established in single word reading studies, however, morphological priming effects in longer texts have not been observed, and, to the best of our knowledge, paradigmatic effects in text reading have not yet been examined. The current study utilized the Multilingual Eye-Tracking Corpus MECO (Siegelman et al., 2022) to explore paradigmatic and morphological priming effects during text reading in Estonian and Finnish, two morphologically rich Finno-Ugric languages. The results showed clear inflectional paradigm size effects for Estonian during text reading in several eye movement measures, but not for Finnish. This may be linked to the support from the inflectional paradigm being semantically more beneficial to the reader in Estonian than in Finnish. The current study also showed clear long-lag inflectional priming effects in text reading, unlike what was observed in prior studies in Dutch, English, and Spanish. This study is thus the first to show that inflectional priming can extend beyond word or sentence level and suggest that inflectional variants of a particular word in Estonian and Finnish get and remain activated even when text context is present.

Pseudo-compound words (e.g., hippie) are words that look like compound words (e.g., snowman) but, in fact, do not have the morphemic structure of a compound word. For instance, the pseudo-compound word hippie has hip and pie embedded in it, but they do not function as morphemes. Pseudo-compound words vary in terms of phonological transparency. Some, such as pumpkin, are phonologically transparent because the pronunciations of pump and kin are maintained when these pseudo-constituents become part of pumpkin. In contrast, the pseudo-compound word carrot is phonologically opaque because the pronunciations of car and rot change when they are embedded in carrot. Previous studies have demonstrated that compound words go through morphological decomposition and attempts at meaning construction during written production tasks. For instance, compound words are not output as single units during typing tasks, but rather are typed in chunks based on their morphology (e.g., snowball is typed in two parts: first as snow and then as ball), which results in an increase in typing latencies at the morpheme boundary (i.e., between the last letter of the first constituent and first letter of the second constituent). The same is true for pseudo-compound words, even though these words do not have the morphemic structure of a compound word. Given that previous research has shown that the morphological decomposition of compound words during typing tasks looks different depending on the semantic transparency of the compound word’s constituents (i.e., the degree to which the meaning of each constituent of a compound word contributes to the word’s overall meaning), we wanted to examine whether the level of phonological transparency of the pseudo-constituents of a pseudo-compound word influences typing latencies at the pseudo-morpheme boundary.

This study investigated the representation of compound words in the mental lexicon by examining compound word production in bilingual speakers with aphasia. Eight bilingual speakers with aphasia named pictures of concepts with either compound or (non-compound) simple names in both of their languages. Error types were coded and analysed within and across languages with a particular focus on constituent and language mixing errors in compound words.

Four participants showed significantly greater accuracy for simple than compound words, three in their dominant and/or more proficient language, and one in their non-dominant language. Constituent errors were observed for all eight bilingual participants during compound word naming, while language mixing errors were observed in six participants.

The observed error patterns support co-activation of (compound) representations during word retrieval in bilingual speakers. Language mixing errors suggest that the bilingual lexicon stores a morphologically structured compound representation, an assumption that is consistent with a multiple-lemma representation of compounds. Further research is required to explore the extent to which constituent-specific access processes are at play in bilingual compound production.