Keynote Presentation

This page shows all conference presentations with the type Keynote Presentation.

Presentations

Cognitive and Ethical Alignment of LLMs with Humans for Writing Research and Instruction

Abstract

Large language models (LLMs) have transformed the study of writing. In linguistics, they catalyzed a shift from the generative grammar paradigm that dominated the latter half of the 20th century. Beyond their practical utility, LLMs provide strong empirical support for connectionist theories of human language processing, showing that complex linguistic behavior can emerge from statistical learning and distributed representations rather than relying solely on (innate) symbolic rules. At the same time, LLMs raise serious questions about alignment with human values, interpretability, and their impact on writing instruction and assessment. Constructing AI systems that simulate human linguistic behavior while aligning with human intentions, reasoning, and values offers both practical and research advantages. This keynote presents two projects that illustrate how LLMs can be aligned with human cognition and ethical principles in writing research and instruction. The first project leverages eye-tracking data, specifically writers' lookback fixations on text produced so far, to guide sentence completions in the emerging text. This approach operationalizes a long-standing hypothesis that writers look back at the text they have produced to support planning of what to say next. By conditioning LLM-based sentence completions on lookback fixation patterns, LLMs produce text that is more closely aligned with a writer's evolving intentions. This work provides empirical evidence for the cognitive function of lookback behavior and establishes a foundation for AI systems for writing support that operate in alignment with human cognitive processes. The second project introduces a hybrid neurosymbolic AI framework for evaluating student argumentative writing. In this framework, LLM-driven inferences from source texts and student essays are constrained by symbolic reasoning that captures ethical norms, logical standards, and pedagogical criteria. By integrating the transparency and reliability of symbolic AI with the flexibility of LLMs in natural language understanding, this approach produces interpretable, robust evaluations of student writing that align with human ethical values. Together, these projects demonstrate that aligning LLM behavior with human cognition and ethical principles can advance both the science of writing and instructional practice. By incorporating cognitive signals and symbolic constraints, AI systems can support and evaluate writing in ways that reflect human intentions, uphold reasoning standards, and promote responsible, interpretable applications of technology.

Bridging Research and Classroom Practice: Improving Writing in Diverse Primary Classrooms

Abstract

Primary school lays the foundation for writing development, where teachers should help students acquire strategies to produce texts independently and foster confidence and motivation in their writing. However, many teachers feel underprepared to support children’s writing beyond spelling, and evidence-based practices often fail to reach the classroom. At the same time, research has shown that many primary children—particularly boys and students whose family language differs from that of instruction—struggle with text production, and writing motivation tends to decline toward the end of primary school. The keynote presents the project KommSchreib! (‘Let’s Write’), funded by the German Federal Ministry of Education and Research, which aimed to address these challenges by bringing evidence-based writing practices into upper primary schools (Grades 3-4). We conducted a two-group quasi-experimental study comparing a business-as-usual control group with an experimental group whose teachers participated in a multi-component professional learning intervention. The intervention focused on evidence-based writing instruction and included teaching materials designed to support student motivation and address diversity. In total, 58 teachers and their 1174 students participated. In addition to the teacher-led intervention in regular classes, project members conducted small-group afternoon writing workshops for at-risk writers with extra practice at tablets, embedded in activities connected to students’ lives. We assessed writing and motivational outcomes, examined treatment fidelity measures, and conducted interviews with participating teachers. The keynote will highlight key findings, including positive effects on students’ writing, explore motivational trajectories, and discuss teacher-related and institutional factors influencing the implementation of evidence-based practices in classrooms.

Hayes Award 2024 Lecture – Writing Fluency in the Perspective of Fluency Research

Abstract

Writing fluency is a fundamental aspect of writing development and is closely related to other 'fluencies', such as reading and speaking fluency. Research on fluency highlights that proficient performance relies on the interaction of automatised, low-level processes, such as transcription in writing or word recognition in reading, and controlled, attention-demanding processes like translating ideas into text or reading with prosody. Over time, writing fluency develops as these fundamental processes become more automated, freeing up cognitive resources for higher-order writing skills such as planning, revising, and producing text strategically. While training programmes targeting fluency can improve these fundamental processes, evidence suggests that isolated practice often has limited impact. Integration into broader literacy instruction is therefore essential for achieving long-lasting results. This keynote will discuss theoretical and developmental perspectives on writing fluency, its connections to reading, speaking and listening fluency, and the implications for designing effective, integrated instructional approaches.