Department of Language Science

Language varieties such as African American English and Southern American English are well-established dialects of English with historic importance and modern cultural capital. While features of language varieties, such as nonmainstream dialects, are well documented in children’s spoken language, less has been studied about child dialect speakers’ writing in academic settings. In this study, we explored the features and frequency of use of African American English and Southern American English in a sample of 250 second- and third-grade students located in the southeastern part of the US. We analyzed and compared nonmainstream dialect production in an oral narrative task and between 2 different written language samples, one elicited by a narrative prompt and the other by an expository prompt. Importantly, we discuss implications for teachers and other practitioners to consider when working with diverse students.

This study investigated the relation between reading comprehension and written composition in Grade 2, as well as how language and literacy skills explain this connection. Additionally, it explored how kindergarten language, cognitive, and literacy skills relate to Grade 2 reading comprehension and written composition through Grade 2 language and literacy skills. The study followed 261 English-speaking students (55% boys) from kindergarten through grade 2 in the US. The sample’s racial composition in kindergarten was 53% white, 33% African American, 3% Hispanic, and 5% mixed race. Grade 2 skills included reading comprehension, written composition, oral discourse skills, lexical literacy skills, and handwriting fluency. Kindergarten assessments covered a broader range of language, cognitive, and literacy skills. Confirmatory factor analysis revealed a strong correlation (.82) between reading comprehension and written composition. Structural equation modeling showed that grade 2 oral discourse language and lexical literacy skills were related to both reading comprehension and writing quality, and accounted for their relation. Handwriting fluency was uniquely related to written composition but not to reading comprehension. Kindergarten skills made substantial indirect contributions to grade 2 reading comprehension and writing quality through grade 2 language and literacy skills. The results indicate that reading comprehension and written composition are strongly related for beginning readers and writers, and support the shared skills hypothesis for their relation. Findings also reveal the nature of the influence of kindergarten skills on grade 2 skills.

This study aimed to enhance our understanding of Spanish-English emergent bilingual children's spelling development by examining the relation between their spelling skills in both languages over time. We also investigated two different scoring methods for spelling and how the language of instruction influenced these relations. The study included 209 Spanish-English bilingual children in the U.S. assessed in Grades 1 and 3. Results showed that children's spelling performance in Spanish and English significantly differed depending on their instructional programs. Using correctness scores, Spanish spelling positively predicted later English spelling, whereas English spelling showed an initial negative relation to future Spanish spelling. This negative relation was not observed when text distance scores were used or when instructional program was explicitly modeled, suggesting the initial negative finding likely reflected measurement limitations in correctness scoring and instructional context rather than negative transfer. Multi-group analyses revealed that most developmental relations did not differ by instructional program. However, Spanish spelling in Grade 1 predicted English spelling in Grade 3 only for children in dual immersion programs when text distance scores were used, highlighting the role of instructional input in cross-linguistic transfer. These findings emphasize the potential benefits of using nonbinary scoring methods for bilingual populations and underscore the importance of considering instructional context when examining bilingual spelling development.

As generative AI becomes ubiquitous, writers must decide if, when, and how to incorporate generative AI into their writing process. Educators must sort through their role in preparing students to make these decisions in a quickly evolving technological landscape. We created an AI-enabled writing tool that provides scaffolded use of a large language model as part of a research study on integrating generative AI into an upper division STEM writing-intensive course. Drawing on decades of research on integrating digital tools into instruction and writing research, we discuss the framework that drove our initial design considerations and instructional resources. We then share our findings from a year of design-based implementation research during the 2023–2024 academic year. Our original instruction framework identified the need for students to understand, access, prompt, corroborate, and incorporate the generative AI use effectively. In this paper, we explain the need for students to think first, before using AI, move through good enough prompting to agentic iterative prompting, and reflect on their use at the end. We also provide emerging best practices for instructors, beginning with identifying learning objectives, determining the appropriate AI role, revising the content, reflecting on the revised curriculum, and reintroducing learning as needed. We end with an indication of our future directions.

Much previous work has shown that sibilant-initial complex onsets (SC onsets) differ in their typological, phonological, articulatory, and acquisitional properties from other onsets. The exact mechanism(s) underlying these differences are poorly understood. In this study, we investigate the acquisition and production of L2 English complex onsets by L1 Farsi speakers, focusing on differences between SC onsets and other onset types. Results from an experimental study corroborate past reports that Farsi speakers repair most SC onsets using epenthesis before the cluster and other onsets using epenthesis into the cluster. The results also support the claim that SC onsets are more difficult to produce and are acquired more slowly than other onsets. A phonological modeling study suggests that the epenthesis asymmetries observed in the experimental study are best accounted for by a pressure to maximize perceptual similarity to the unepenthesized forms. We close with speculation on how the variegated behavior of SC onsets can be given a holistic explanation under a perceptual account.

This study compared the processing of non-binary morphemes in Spanish (e.g., todxs, todes) with the processing of canonical grammatical gender violations in Spanish pronouns (e.g., Los maestros… todas…). Using self-paced reading, the study examined how individual differences in working memory and gender/sex diversity beliefs affected language processing at three regions of interest (ROI): the pronoun, the pronoun +1, and the pronoun +2. Seventy-eight Spanish-English bilinguals completed two self-paced reading tasks, one with non-binary pronouns and another with grammatical gender violations, as well as a working memory task, a language dominance questionnaire, and a gender/sex diversity beliefs questionnaire. Processing costs were operationalized as longer reaction times (RTs) or inaccurate responses. Results showed overall processing costs for non-binary morphemes at all 3 ROIs, but no processing costs were observed in terms of accuracy or response times to the comprehension question. The results suggest that processing non-binary pronouns results in a small processing cost that does not affect overall sentence comprehension. The small observed processing cost was moderated by gender/sex diversity beliefs, with gender normative beliefs increasing RTs at the pronoun and affirmation of diverse gender identities beliefs reducing the RTs at the second spillover region. In contrast, grammatical gender violations only showed a processing cost at the first spillover region and were not moderated by working memory nor gender/sex diversity beliefs. Taken together, the results suggest that non-binary pronouns are processed differently than grammatical gender violations and that the small processing cost they impose can lead to good enough comprehension.

INTRODUCTION: This mixed-method longitudinal study examined American adolescents' meaning making of salient COVID-19 pandemic events. METHOD: Within phone interviews, adolescents (N = 124, M_age = 15.76 years; 46% Latine) narrated their most emotionally impactful pandemic experience at two time points ~30 days apart between July 2020 and March 2021. Narratives were coded for (1) content (i.e., event-type, relation to the pandemic, and the valence of the event [positive or negative]), (2) linguistic markers of subjective event processing (internal state language such as positive emotion, negative emotion, and cognition words), (3) narrative meaning-making, and (4) the outcome of adolescents' meaning-making (i.e., their "meanings made"). RESULTS: About 30% of adolescents spontaneously made meaning of their experience. Negative emotion words within narratives at time 1 positively predicted meaning making at time 2. Meaning making at time 1 predicted increased use of cognition words at time 2. Meaning making themes included: recognizing the threat of COVID-19, coping with a pandemic, and shifts in perspectives. DISCUSSION: Salient emotional experiences that occur during adolescence are likely to be remembered and contribute to one's life story. This work provides a window into how the COVID-19 pandemic may have shaped adolescent development in the United States.

Humans can modify their speech to improve intelligibility in noisy environments. With the advancement of speech synthesis technology, machines may also synthesize voices that remain highly intelligible in noise condition. This study evaluates both the subjective and objective intelligibility of synthesized speech in speech-shaped noise from three major speech synthesis platforms. It was found that synthesized voices have a similar intelligibility range to human voices, and some synthesized voices were more intelligible than human voices. It was also found that two modern automatic speech recognition systems recognized 10% more words than human listeners.

OBJECTIVES: Cochlear implants restore functional hearing but may cause side effects like facial nerve stimulation, sound sensitivity or reactive tinnitus. The present study aimed to establish a general framework for optimizing stimulation parameters to manage these side effects while maximizing speech perception performance. A second objective was to understand how side effect origins impact treatment outcomes. METHODS: Eight adult cochlear implant subjects had intolerable side effects that rendered device usage difficult or even impossible. New maps were created by reducing stimulation levels, increasing pulse duration, reducing stimulation rate, altering channel gains and frequency maps, deactivating problematic electrodes, or a combination of the above. Outcomes were measured in terms of side effect reduction and changes in speech performance. RESULTS: Facial nerve stimulation was reduced or eliminated in five of five subjects. Sound hypersensitivity was eliminated in two of two subjects. Tinnitus was alleviated in three of four subjects, while the remaining one with cerebellar malformation experienced no change. Speech performance was either maintained or improved in all subjects. Except for the subject with cerebellar malformation who chose to explant the device, all subjects were able to use the implant effectively without bothersome side effects. DISCUSSION: Facial nerve stimulation is usually related to electric current spread on the same side, which can be effectively managed by customized strategies. In contrast, the origins of sound sensitivity and reactive tinnitus are more variable and likely more difficult to manage. CONCLUSION: Customized mapping can alleviate cochlear implant side effects without compromising speech performance.

Large language models (LLMs) require alignment to effectively and safely follow user instructions. This process necessitates training an aligned version for every base model, resulting in significant computational overhead. In this work, we propose NUDGING, a simple, training-free algorithm that aligns any base model at inference time using a small aligned model. NUDGING is motivated by recent findings that alignment primarily alters the model's behavior on a small subset of stylistic tokens (e.g., discourse markers). We find that base models are significantly more uncertain when generating these tokens. Building on this insight, NUDGING employs a small aligned model to generate nudging tokens to guide the base model's output during decoding when the base model's uncertainty is high, with only a minor additional inference overhead. We evaluate NUDGING across 3 model families on a diverse range of open-instruction tasks. Without any training, nudging a large base model with a 7×-14× smaller aligned model achieves zero-shot performance comparable to, and sometimes surpassing, that of large aligned models. By operating at the token level, NUDGING enables off-the-shelf collaboration between model families. For instance, nudging Gemma-2-27b with Llama-27b-chat outperforms Llama-2-70b-chat on various tasks. Overall, our work offers a modular and cost-efficient solution to LLM alignment. Our code and demo are available at: https://fywalter.github.io/nudging/.