Nura Aljaafari

Nura Aljaafari, Danilo S. Carvalho, André Freitas

Locating and editing knowledge in large language models (LLMs) is crucial for enhancing their accuracy, safety, and inference rationale. We introduce ``concept editing'', an innovative variation of knowledge editing that uncovers conceptualisation mechanisms within these models. Using the reverse dictionary task, inference tracing, and input abstraction, we analyse the Multi-Layer Perceptron (MLP), Multi-Head Attention (MHA), and hidden state components of transformer models. Our results reveal distinct patterns: MLP layers employ key-value retrieval mechanism and context-dependent processing, which are highly associated with relative input tokens. MHA layers demonstrate a distributed nature with significant higher-level activations, suggesting sophisticated semantic integration. Hidden states emphasise the importance of the last token and top layers in the inference process. We observe evidence of gradual information building and distributed representation. These observations elucidate how transformer models process semantic information, paving the way for targeted interventions and improved interpretability techniques. Our work highlights the complex, layered nature of semantic processing in LLMs and the challenges of isolating and modifying specific concepts within these models.

Nura Aljaafari, Danilo S. Carvalho, André Freitas

Understanding the internal mechanisms of large language models (LLMs) is integral to enhancing their reliability, interpretability, and inference processes. We present Constituent-Aware Pooling (CAP), a methodology designed to analyse how LLMs process compositional linguistic structures. Grounded in principles of compositionality, mechanistic interpretability, and information theory, CAP systematically intervenes in model activations through constituent-based pooling at various model levels. Our experiments on inverse definition modelling, hypernym and synonym prediction reveal critical insights into transformers' limitations in handling compositional abstractions. No specific layer integrates tokens into unified semantic representations based on their constituent parts. We observe fragmented information processing, which intensifies with model size, suggesting that larger models struggle more with these interventions and exhibit greater information dispersion. This fragmentation likely stems from transformers' training objectives and architectural design, preventing systematic and cohesive representations. Our findings highlight fundamental limitations in current transformer architectures regarding compositional semantics processing and model interpretability, underscoring the critical need for novel approaches in LLM design to address these challenges.

Nura Aljaafari, Danilo S. Carvalho, André Freitas

Large language models (LLMs) struggle with compositional generalisation, limiting their ability to systematically combine learned components to interpret novel inputs. While architectural modifications, fine-tuning, and data augmentation improve compositionality, they often have limited adaptability, face scalability constraints, or yield diminishing returns on real data. To address this, we propose CARMA, an intervention that enhances the stability and robustness of compositional reasoning in LLMs while preserving fine-tuned performance. CARMA employs mutual information regularisation and layer-wise stability constraints to mitigate feature fragmentation, ensuring structured representations persist across and within layers. We evaluate CARMA on inverse dictionary modelling and sentiment classification, measuring its impact on semantic consistency, performance stability, and robustness to lexical perturbations. Results show that CARMA reduces the variability introduced by fine-tuning, stabilises token representations, and improves compositional reasoning. While its effectiveness varies across architectures, CARMA's key strength lies in reinforcing learned structures rather than introducing new capabilities, making it a scalable auxiliary method. These findings suggest that integrating CARMA with fine-tuning can improve compositional generalisation while maintaining task-specific performance in LLMs.

Nura Aljaafari, Danilo S. Carvalho, Andre Freitas

Mechanistic interpretability produces circuit-level causal analyses of neural network behaviour, but discovered circuits often remain isolated experimental artefacts: there is no shared formal representation for what circuits compute, how they relate, or when two findings provide evidence for the same mechanism. This work provides a formal infrastructure for cumulative mechanistic science by treating circuit interpretation as inductive theory construction. Each circuit is characterised at two levels: a Causal Functional Signature (CFS), which grounds component behaviour in causal attribution evidence and token role profiles, and an architectural signature $τ_{\mathrm{arch}}$, learned by inductive logic programming (ILP) from scale-invariant structural predicates. Together, these constitute a formal coherence layer that makes mechanistic claims explicit, comparable via $θ$-subsumption, and portable across model scales. CFS reveals qualitatively distinct computational strategies across task types, including attention-mediated copying versus MLP-mediated binding. ILP signatures achieve substantially better structural separation than graph kernel and feature-vector baselines, and support principled transfer across model scales and architecture families.

Nura Aljaafari, Danilo S. Carvalho, André Freitas

Understanding when and how linguistic knowledge emerges during language model training remains a central challenge for interpretability. Most existing tools are post hoc, rely on scalar metrics, or require nontrivial integration effort, making comprehensive interpretability analysis difficult to deploy and maintain. We introduce TRACE, a modular toolkit for training and inference-time interpretability analysis of transformer models. It enables lightweight, in-training analysis of linguistic and representational signals, including features probing, intrinsic dimensionality, Hessian curvature, and output diagnostics. It integrates with ABSynth, a controllable synthetic corpus generator that provides structured annotations for precise evaluation of linguistic feature acquisition. Experiments with autoregressive transformers demonstrate that TRACE reveals developmental phenomena such as early syntactic emergence, delayed semantic acquisition, and representational compression, signals overlooked by traditional scalar metrics such as loss or accuracy. With minimal integration effort, the tool enables layer-wise diagnostics, convergence-based early stopping, and detection of structural errors, making transformer analysis interpretable, actionable, and reproducible.