Mario Giulianelli

Aarohi Srivastava, Abhinav Rastogi, Abhishek Rao et al. · allen-ai, amazon-science

Language models demonstrate both quantitative improvement and new qualitative capabilities with increasing scale. Despite their potentially transformative impact, these new capabilities are as yet poorly characterized. In order to inform future research, prepare for disruptive new model capabilities, and ameliorate socially harmful effects, it is vital that we understand the present and near-future capabilities and limitations of language models. To address this challenge, we introduce the Beyond the Imitation Game benchmark (BIG-bench). BIG-bench currently consists of 204 tasks, contributed by 450 authors across 132 institutions. Task topics are diverse, drawing problems from linguistics, childhood development, math, common-sense reasoning, biology, physics, social bias, software development, and beyond. BIG-bench focuses on tasks that are believed to be beyond the capabilities of current language models. We evaluate the behavior of OpenAI's GPT models, Google-internal dense transformer architectures, and Switch-style sparse transformers on BIG-bench, across model sizes spanning millions to hundreds of billions of parameters. In addition, a team of human expert raters performed all tasks in order to provide a strong baseline. Findings include: model performance and calibration both improve with scale, but are poor in absolute terms (and when compared with rater performance); performance is remarkably similar across model classes, though with benefits from sparsity; tasks that improve gradually and predictably commonly involve a large knowledge or memorization component, whereas tasks that exhibit "breakthrough" behavior at a critical scale often involve multiple steps or components, or brittle metrics; social bias typically increases with scale in settings with ambiguous context, but this can be improved with prompting.

Dieuwke Hupkes, Mario Giulianelli, Verna Dankers et al. · amazon-science, cambridge

The ability to generalise well is one of the primary desiderata of natural language processing (NLP). Yet, what 'good generalisation' entails and how it should be evaluated is not well understood, nor are there any evaluation standards for generalisation. In this paper, we lay the groundwork to address both of these issues. We present a taxonomy for characterising and understanding generalisation research in NLP. Our taxonomy is based on an extensive literature review of generalisation research, and contains five axes along which studies can differ: their main motivation, the type of generalisation they investigate, the type of data shift they consider, the source of this data shift, and the locus of the shift within the modelling pipeline. We use our taxonomy to classify over 400 papers that test generalisation, for a total of more than 600 individual experiments. Considering the results of this review, we present an in-depth analysis that maps out the current state of generalisation research in NLP, and we make recommendations for which areas might deserve attention in the future. Along with this paper, we release a webpage where the results of our review can be dynamically explored, and which we intend to update as new NLP generalisation studies are published. With this work, we aim to take steps towards making state-of-the-art generalisation testing the new status quo in NLP.

Mario Giulianelli, Andreas Opedal, Ryan Cotterell

We introduce a generalization of classic information-theoretic measures of predictive uncertainty in online language processing, based on the simulation of expected continuations of incremental linguistic contexts. Our framework provides a formal definition of anticipatory and responsive measures, and it equips experimenters with the tools to define new, more expressive measures beyond standard next-symbol entropy and surprisal. While extracting these standard quantities from language models is convenient, we demonstrate that using Monte Carlo simulation to estimate alternative responsive and anticipatory measures pays off empirically: New special cases of our generalized formula exhibit enhanced predictive power compared to surprisal for human cloze completion probability as well as ELAN, LAN, and N400 amplitudes, and greater complementarity with surprisal in predicting reading times.

Mario Giulianelli, Sarenne Wallbridge, Raquel Fernández

We present information value, a measure which quantifies the predictability of an utterance relative to a set of plausible alternatives. We introduce a method to obtain interpretable estimates of information value using neural text generators, and exploit their psychometric predictive power to investigate the dimensions of predictability that drive human comprehension behaviour. Information value is a stronger predictor of utterance acceptability in written and spoken dialogue than aggregates of token-level surprisal and it is complementary to surprisal for predicting eye-tracked reading times.

Mario Giulianelli, Andrey Kutuzov, Lidia Pivovarova

Morphological and syntactic changes in word usage (as captured, e.g., by grammatical profiles) have been shown to be good predictors of a word's meaning change. In this work, we explore whether large pre-trained contextualised language models, a common tool for lexical semantic change detection, are sensitive to such morphosyntactic changes. To this end, we first compare the performance of grammatical profiles against that of a multilingual neural language model (XLM-R) on 10 datasets, covering 7 languages, and then combine the two approaches in ensembles to assess their complementarity. Our results show that ensembling grammatical profiles with XLM-R improves semantic change detection performance for most datasets and languages. This indicates that language models do not fully cover the fine-grained morphological and syntactic signals that are explicitly represented in grammatical profiles. An interesting exception are the test sets where the time spans under analysis are much longer than the time gap between them (for example, century-long spans with a one-year gap between them). Morphosyntactic change is slow so grammatical profiles do not detect in such cases. In contrast, language models, thanks to their access to lexical information, are able to detect fast topical changes.

Mario Giulianelli, Arabella Sinclair, Raquel Fernández

Speakers repeat constructions frequently in dialogue. Due to their peculiar information-theoretic properties, repetitions can be thought of as a strategy for cost-effective communication. In this study, we focus on the repetition of lexicalised constructions -- i.e., recurring multi-word units -- in English open-domain spoken dialogues. We hypothesise that speakers use construction repetition to mitigate information rate, leading to an overall decrease in utterance information content over the course of a dialogue. We conduct a quantitative analysis, measuring the information content of constructions and that of their containing utterances, estimating information content with an adaptive neural language model. We observe that construction usage lowers the information content of utterances. This facilitating effect (i) increases throughout dialogues, (ii) is boosted by repetition, (iii) grows as a function of repetition frequency and density, and (iv) is stronger for repetitions of referential constructions.

Mario Giulianelli

This position paper proposes a conceptual framework for the design of Natural Language Generation (NLG) systems that follow efficient and effective production strategies in order to achieve complex communicative goals. In this general framework, efficiency is characterised as the parsimonious regulation of production and comprehension costs while effectiveness is measured with respect to task-oriented and contextually grounded communicative goals. We provide concrete suggestions for the estimation of goals, costs, and utility via modern statistical methods, demonstrating applications of our framework to the classic pragmatic task of visually grounded referential games and to abstractive text summarisation, two popular generation tasks with real-world applications. In sum, we advocate for the development of NLG systems that learn to make pragmatic production decisions from experience, by reasoning about goals, costs, and utility in a human-like way.

Matteo Gay, Coleman Haley, Mario Giulianelli et al.

The Uniform Information Density (UID) hypothesis posits that speakers are subject to a communicative pressure to distribute information evenly within utterances, minimising surprisal variance. While this hypothesis has been tested empirically, prior studies are limited exclusively to text-only inputs, abstracting away from the perceptual context in which utterances are produced. In this work, we present the first computational study of UID in visually grounded settings. We estimate surprisal using multilingual vision-and-language models over image-caption data in 30 languages and visual storytelling data in 13 languages, together spanning 11 families. We find that grounding on perception consistently smooths the distribution of information, increasing both global and local uniformity across typologically diverse languages compared to text-only settings. In visual narratives, grounding in both image and discourse contexts has additional effects, with the strongest surprisal reductions occurring at the onset of discourse units. Overall, this study takes a first step towards modelling the temporal dynamics of information flow in ecologically plausible, multimodal language use, and finds that grounded language exhibits greater information uniformity, supporting a context-sensitive formulation of UID.

Aron Molnar, Jaap Jumelet, Mario Giulianelli et al.

Language models are often used as the backbone of modern dialogue systems. These models are pre-trained on large amounts of written fluent language. Repetition is typically penalised when evaluating language model generations. However, it is a key component of dialogue. Humans use local and partner specific repetitions; these are preferred by human users and lead to more successful communication in dialogue. In this study, we evaluate (a) whether language models produce human-like levels of repetition in dialogue, and (b) what are the processing mechanisms related to lexical re-use they use during comprehension. We believe that such joint analysis of model production and comprehension behaviour can inform the development of cognitively inspired dialogue generation systems.

Eleftheria Tsipidi, Samuel Kiegeland, Francesco Ignazio Re et al.

Probing has shown that language model representations encode rich linguistic information, but it remains unclear whether they also capture cognitive signals about human processing. In this work, we probe language model representations for human reading times. Using regularized linear regression on two eye-tracking corpora spanning five languages (English, Greek, Hebrew, Russian, and Turkish), we compare the representations from every model layer against scalar predictors -- surprisal, information value, and logit-lens surprisal. We find that the representations from early layers outperform surprisal in predicting early-pass measures such as first fixation and gaze duration. The concentration of predictive power in the early layers suggests that human-like processing signatures are captured by low-level structural or lexical representations, pointing to a functional alignment between model depth and the temporal stages of human reading. In contrast, for late-pass measures such as total reading time, scalar surprisal remains superior, despite its being a much more compressed representation. We also observe performance gains when using both surprisal and early-layer representations. Overall, we find that the best-performing predictor varies strongly depending on the language and eye-tracking measure.

Mario Giulianelli, Joris Baan, Wilker Aziz et al.

In Natural Language Generation (NLG) tasks, for any input, multiple communicative goals are plausible, and any goal can be put into words, or produced, in multiple ways. We characterise the extent to which human production varies lexically, syntactically, and semantically across four NLG tasks, connecting human production variability to aleatoric or data uncertainty. We then inspect the space of output strings shaped by a generation system's predicted probability distribution and decoding algorithm to probe its uncertainty. For each test input, we measure the generator's calibration to human production variability. Following this instance-level approach, we analyse NLG models and decoding strategies, demonstrating that probing a generator with multiple samples and, when possible, multiple references, provides the level of detail necessary to gain understanding of a model's representation of uncertainty. Code available at https://github.com/dmg-illc/nlg-uncertainty-probes.

Raghu Arghal, Fade Chen, Niall Dalton et al.

Understanding an agent's goals helps explain and predict its behaviour, yet there is no established methodology for reliably attributing goals to agentic systems. We propose a framework for evaluating goal-directedness that integrates behavioural evaluation with interpretability-based analyses of models' internal representations. As a case study, we examine an LLM agent navigating a 2D grid world toward a goal state. Behaviourally, we evaluate the agent against an optimal policy across varying grid sizes, obstacle densities, and goal structures, finding that performance scales with task difficulty while remaining robust to difficulty-preserving transformations and complex goal structures. We then use probing methods to decode the agent's internal representations of the environment state and its multi-step action plans. We find that the LLM agent non-linearly encodes a coarse spatial map of the environment, preserving approximate task-relevant cues about its position and the goal location; that its actions are broadly consistent with these internal representations; and that reasoning reorganises them, shifting from broader environment structural cues toward information supporting immediate action selection. Our findings support the view that introspective examination is required beyond behavioural evaluations to characterise how agents represent and pursue their objectives.

Ivar Frisch, Mario Giulianelli

While both agent interaction and personalisation are vibrant topics in research on large language models (LLMs), there has been limited focus on the effect of language interaction on the behaviour of persona-conditioned LLM agents. Such an endeavour is important to ensure that agents remain consistent to their assigned traits yet are able to engage in open, naturalistic dialogues. In our experiments, we condition GPT-3.5 on personality profiles through prompting and create a two-group population of LLM agents using a simple variability-inducing sampling algorithm. We then administer personality tests and submit the agents to a collaborative writing task, finding that different profiles exhibit different degrees of personality consistency and linguistic alignment to their conversational partners. Our study seeks to lay the groundwork for better understanding of dialogue-based interaction between LLMs and highlights the need for new approaches to crafting robust, more human-like LLM personas for interactive environments.

Tom Utting, Mario Giulianelli, Arabella Sinclair

We model utterance production as probabilistic cost-sensitive choice over contextual alternatives, using information-theoretic notions of cost. We distinguish between goal-directed alternatives that realise a fixed communicative intent and goal-agnostic alternatives defined only by contextual plausibility, allowing us to derive speaker- and listener-oriented interpretations of different cost measures. We present a procedure to generate both types of alternative sets using language models. Analysing production choices in open-ended dialogue under both deterministic and probabilistic cost minimisation, we find that surprisal minimisation relative to goal-directed alternatives provides the strongest predictive account under both analyses. By contrast, uniform information density and length-based costs exhibit weaker and less consistent predictive power across conditions. More broadly, our study suggests that alternative-conditioned optimisation with LM-generated alternatives provides a principled framework for studying speaker and listener pressures in naturalistic language production.

Tim Vieira, Ben LeBrun, Mario Giulianelli et al.

Modern language models are internally -- and mathematically -- distributions over $\it{token}$ strings rather than $\it{character}$ strings, posing numerous challenges for programmers building user applications on top of them. For example, if a prompt is specified as a character string, it must be tokenized before passing it to the token-level language model. Thus, the tokenizer and consequent processing are very sensitive to the specification of the prompt (e.g., whether the prompt ends with a space or not). This paper presents algorithms for converting token-level language models to character-level ones. We present both exact and approximate algorithms. In the empirical portion of the paper, we benchmark the practical runtime and approximation quality. Across four publicly available language models, we find that -- even with a small computation budget -- our method is able to accurately approximate the character-level distribution at reasonably fast speeds, and that a significant improvement in the language model's compression rate (bits/byte) is achieved.

Gonçalo Guiomar, Elia Torre, Pehuen Moure et al.

Do language models make decisions under uncertainty like humans do, and what role does chain-of-thought (CoT) reasoning play in the underlying decision process? We introduce an active probabilistic reasoning task that cleanly separates sampling (actively acquiring evidence) from inference (integrating evidence toward a decision). Benchmarking humans and a broad set of contemporary large language models against near-optimal reference policies reveals a consistent pattern: extended reasoning is the key determinant of strong performance, driving large gains in inference and producing belief trajectories that become strikingly human-like, while yielding only modest improvements in active sampling. To explain these differences, we fit a mechanistic model that captures systematic deviations from optimal behavior via four interpretable latent variables: memory, strategy, choice bias, and occlusion awareness. This model places humans and models in a shared low-dimensional cognitive space, reproduces behavioral signatures across agents, and shows how chain-of-thought shifts language models toward human-like regimes of evidence accumulation and belief-to-choice mapping, tightening alignment in inference while leaving a persistent gap in information acquisition.

Eleftheria Tsipidi, Franz Nowak, Ryan Cotterell et al.

The Uniform Information Density (UID) hypothesis posits that speakers tend to distribute information evenly across linguistic units to achieve efficient communication. Of course, information rate in texts and discourses is not perfectly uniform. While these fluctuations can be viewed as theoretically uninteresting noise on top of a uniform target, another explanation is that UID is not the only functional pressure regulating information content in a language. Speakers may also seek to maintain interest, adhere to writing conventions, and build compelling arguments. In this paper, we propose one such functional pressure; namely that speakers modulate information rate based on location within a hierarchically-structured model of discourse. We term this the Structured Context Hypothesis and test it by predicting the surprisal contours of naturally occurring discourses extracted from large language models using predictors derived from discourse structure. We find that hierarchical predictors are significant predictors of a discourse's information contour and that deeply nested hierarchical predictors are more predictive than shallow ones. This work takes an initial step beyond UID to propose testable hypotheses for why the information rate fluctuates in predictable ways

Clara Meister, Mario Giulianelli, Tiago Pimentel · cambridge

Surprisal theory posits that the cognitive effort required to comprehend a word is determined by its contextual predictability, quantified as surprisal. Traditionally, surprisal theory treats words as distinct entities, overlooking any potential similarity between them. Giulianelli et al. (2023) address this limitation by introducing information value, a measure of predictability designed to account for similarities between communicative units. Our work leverages Ricotta and Szeidl's (2006) diversity index to extend surprisal into a metric that we term similarity-adjusted surprisal, exposing a mathematical relationship between surprisal and information value. Similarity-adjusted surprisal aligns with information value when considering graded similarities and reduces to standard surprisal when words are treated as distinct. Experimental results with reading time data indicate that similarity-adjusted surprisal adds predictive power beyond standard surprisal for certain datasets, suggesting it serves as a complementary measure of comprehension effort.

Yuxuan Zhu, Tengjun Jin, Yada Pruksachatkun et al.

Benchmarks are essential for quantitatively tracking progress in AI. As AI agents become increasingly capable, researchers and practitioners have introduced agentic benchmarks to evaluate agents on complex, real-world tasks. These benchmarks typically measure agent capabilities by evaluating task outcomes via specific reward designs. However, we show that many agentic benchmarks have issues in task setup or reward design. For example, SWE-bench Verified uses insufficient test cases, while TAU-bench counts empty responses as successful. Such issues can lead to under- or overestimation of agents' performance by up to 100% in relative terms. To make agentic evaluation rigorous, we introduce the Agentic Benchmark Checklist (ABC), a set of guidelines that we synthesized from our benchmark-building experience, a survey of best practices, and previously reported issues. When applied to CVE-Bench, a benchmark with a particularly complex evaluation design, ABC reduces the performance overestimation by 33%.

Tim Vieira, Tianyu Liu, Clemente Pasti et al.

Modern language models represent probability distributions over character strings as distributions over (shorter) token strings derived via a deterministic tokenizer, such as byte-pair encoding. While this approach is highly effective at scaling up language models to large corpora, its current incarnations have a concerning property: the model assigns nonzero probability mass to an exponential number of $\it{noncanonical}$ token encodings of each character string -- these are token strings that decode to valid character strings but are impossible under the deterministic tokenizer (i.e., they will never be seen in any training corpus, no matter how large). This misallocation is both erroneous, as noncanonical strings never appear in training data, and wasteful, diverting probability mass away from plausible outputs. These are avoidable mistakes! In this work, we propose methods to enforce canonicality in token-level language models, ensuring that only canonical token strings are assigned positive probability. We present two approaches: (1) canonicality by conditioning, leveraging test-time inference strategies without additional training, and (2) canonicality by construction, a model parameterization that guarantees canonical outputs but requires training. We demonstrate that fixing canonicality mistakes improves the likelihood of held-out data for several models and corpora.

Filippo Momentè, Alessandro Suglia, Mario Giulianelli et al.

We examine three evaluation paradigms: standard benchmarks (e.g., MMLU and BBH), interactive games (e.g., Signalling Games or Taboo), and cognitive tests (e.g., for working memory or theory of mind). First, we investigate which of the former two-benchmarks or games-is most effective at discriminating LLMs of varying quality. Then, inspired by human cognitive assessments, we compile a suite of targeted tests that measure cognitive abilities deemed essential for effective language use, and we investigate their correlation with model performance in benchmarks and games. Our analyses reveal that interactive games are superior to standard benchmarks in discriminating models. Causal and logical reasoning correlate with both static and interactive tests, while differences emerge regarding core executive functions and social/emotional skills, which correlate more with games. We advocate for the development of new interactive benchmarks and targeted cognitive tasks inspired by assessing human abilities but designed specifically for LLMs.

Magda Dubois, Harry Coppock, Mario Giulianelli et al.

The evaluation of large language model (LLM) outputs is increasingly performed by other LLMs, a setup commonly known as "LLM-as-a-judge", or autograders. While autograders offer a scalable alternative to human evaluation, they have shown mixed reliability and may exhibit systematic biases, depending on response type, scoring methodology, domain specificity, or other factors. Here we propose a statistical framework based on Bayesian generalised linear models (GLMs) that enables researchers to simultaneously assess their autograders while addressing their primary research questions (e.g., LLM evaluation). Our approach models evaluation outcomes (e.g., scores or pairwise preferences) as a function of properties of the grader (e.g., human vs. autograder) and the evaluated item (e.g., response length or the LLM that generated it), allowing for explicit quantification of scoring differences and potential biases within a unified framework. In addition, our method can be used to augment traditional metrics such as inter-rater agreement, by providing uncertainty estimates and clarifying sources of disagreement. Overall, this approach contributes to more robust and interpretable use of autograders in LLM evaluation, enabling both performance analysis and bias detection.

Christopher Summerfield, Lennart Luettgau, Magda Dubois et al.

We examine recent research that asks whether current AI systems may be developing a capacity for "scheming" (covertly and strategically pursuing misaligned goals). We compare current research practices in this field to those adopted in the 1970s to test whether non-human primates could master natural language. We argue that there are lessons to be learned from that historical research endeavour, which was characterised by an overattribution of human traits to other agents, an excessive reliance on anecdote and descriptive analysis, and a failure to articulate a strong theoretical framework for the research. We recommend that research into AI scheming actively seeks to avoid these pitfalls. We outline some concrete steps that can be taken for this research programme to advance in a productive and scientifically rigorous fashion.

Nicola Horst, Davide Mazzaccara, Antonia Schmidt et al.

Interaction between learner and feedback-giver has come into focus recently for post-training of Large Language Models (LLMs), through the use of reward models that judge the appropriateness of a model's response. In this paper, we investigate whether Dialogue Games -- goal-directed and rule-governed activities driven predominantly by verbal actions -- can also serve as a source of feedback signals for learning. We introduce Playpen, an environment for off- and online learning through Dialogue Game self-play, and investigate a representative set of post-training methods: supervised fine-tuning; direct alignment (DPO); and reinforcement learning with GRPO. We experiment with post-training a small LLM (Llama-3.1-8B-Instruct), evaluating performance on unseen instances of training games as well as unseen games, and on standard benchmarks. We find that imitation learning through SFT improves performance on unseen instances, but negatively impacts other skills, while interactive learning with GRPO shows balanced improvements without loss of skills. We release the framework and the baseline training setups to foster research in the promising new direction of learning in (synthetic) interaction.

Eleftheria Tsipidi, Samuel Kiegeland, Franz Nowak et al.

The uniform information density (UID) hypothesis proposes that speakers aim to distribute information evenly throughout a text, balancing production effort and listener comprehension difficulty. However, language typically does not maintain a strictly uniform information rate; instead, it fluctuates around a global average. These fluctuations are often explained by factors such as syntactic constraints, stylistic choices, or audience design. In this work, we explore an alternative perspective: that these fluctuations may be influenced by an implicit linguistic pressure towards periodicity, where the information rate oscillates at regular intervals, potentially across multiple frequencies simultaneously. We apply harmonic regression and introduce a novel extension called time scaling to detect and test for such periodicity in information contours. Analyzing texts in English, Spanish, German, Dutch, Basque, and Brazilian Portuguese, we find consistent evidence of periodic patterns in information rate. Many dominant frequencies align with discourse structure, suggesting these oscillations reflect meaningful linguistic organization. Beyond highlighting the connection between information rate and discourse structure, our approach offers a general framework for uncovering structural pressures at various levels of linguistic granularity.

Bryan Eikema, Anna Rutkiewicz, Mario Giulianelli

Minimum Bayes Risk (MBR) decoding has seen renewed interest as an alternative to traditional generation strategies. While MBR has proven effective in machine translation, where the variability of a language model's outcome space is naturally constrained, it may face challenges in more open-ended tasks such as dialogue or instruction-following. We hypothesise that in such settings, applying MBR with standard similarity-based utility functions may result in selecting responses that are broadly representative of the model's distribution, yet sub-optimal with respect to any particular grouping of generations that share an underlying latent structure. In this work, we introduce three lightweight adaptations to the utility function, designed to make MBR more sensitive to structural variability in the outcome space. To test our hypothesis, we curate a dataset capturing three representative types of latent structure: dialogue act, emotion, and response structure (e.g., a sentence, a paragraph, or a list). We further propose two metrics to evaluate the structural optimality of MBR. Our analysis demonstrates that common similarity-based utility functions fall short by these metrics. In contrast, our proposed adaptations considerably improve structural optimality. Finally, we evaluate our approaches on real-world instruction-following benchmarks, AlpacaEval and MT-Bench, and show that increased structural sensitivity improves generation quality by up to 13.7 percentage points in win rate.

Francesco Ignazio Re, Andreas Opedal, Glib Manaiev et al.

Reading is a process that unfolds across space and time, alternating between fixations where a reader focuses on a specific point in space, and saccades where a reader rapidly shifts their focus to a new point. An ansatz of psycholinguistics is that modeling a reader's fixations and saccades yields insight into their online sentence processing. However, standard approaches to such modeling rely on aggregated eye-tracking measurements and models that impose strong assumptions, ignoring much of the spatio-temporal dynamics that occur during reading. In this paper, we propose a more general probabilistic model of reading behavior, based on a marked spatio-temporal point process, that captures not only how long fixations last, but also where they land in space and when they take place in time. The saccades are modeled using a Hawkes process, which captures how each fixation excites the probability of a new fixation occurring near it in time and space. The duration time of fixation events is modeled as a function of fixation-specific predictors convolved across time, thus capturing spillover effects. Empirically, our Hawkes process model exhibits a better fit to human saccades than baselines. With respect to fixation durations, we observe that incorporating contextual surprisal as a predictor results in only a marginal improvement in the model's predictive accuracy. This finding suggests that surprisal theory struggles to explain fine-grained eye movements.

Taiga Someya, Anej Svete, Brian DuSell et al. · allen-ai, eth-zurich

Inductive biases are inherent in every machine learning system, shaping how models generalize from finite data. In the case of neural language models (LMs), debates persist as to whether these biases align with or diverge from human processing constraints. To address this issue, we propose a quantitative framework that allows for controlled investigations into the nature of these biases. Within our framework, we introduce $m$-local entropy$\unicode{x2013}$an information-theoretic measure derived from average lossy-context surprisal$\unicode{x2013}$that captures the local uncertainty of a language by quantifying how effectively the $m-1$ preceding symbols disambiguate the next symbol. In experiments on both perturbed natural language corpora and languages defined by probabilistic finite-state automata (PFSAs), we show that languages with higher $m$-local entropy are more difficult for Transformer and LSTM LMs to learn. These results suggest that neural LMs, much like humans, are highly sensitive to the local statistical structure of a language.

Anna Bavaresco, Raffaella Bernardi, Leonardo Bertolazzi et al.

There is an increasing trend towards evaluating NLP models with LLMs instead of human judgments, raising questions about the validity of these evaluations, as well as their reproducibility in the case of proprietary models. We provide JUDGE-BENCH, an extensible collection of 20 NLP datasets with human annotations covering a broad range of evaluated properties and types of data, and comprehensively evaluate 11 current LLMs, covering both open-weight and proprietary models, for their ability to replicate the annotations. Our evaluations show substantial variance across models and datasets. Models are reliable evaluators on some tasks, but overall display substantial variability depending on the property being evaluated, the expertise level of the human judges, and whether the language is human or model-generated. We conclude that LLMs should be carefully validated against human judgments before being used as evaluators.

Ece Takmaz, Nicolo' Brandizzi, Mario Giulianelli et al.

Dialogue participants may have varying levels of knowledge about the topic under discussion. In such cases, it is essential for speakers to adapt their utterances by taking their audience into account. Yet, it is an open question how such adaptation can be modelled in computational agents. In this paper, we model a visually grounded referential game between a knowledgeable speaker and a listener with more limited visual and linguistic experience. Inspired by psycholinguistic theories, we endow our speaker with the ability to adapt its referring expressions via a simulation module that monitors the effectiveness of planned utterances from the listener's perspective. We propose an adaptation mechanism building on plug-and-play approaches to controlled language generation, where utterance generation is steered on the fly by the simulator without finetuning the speaker's underlying language model. Our results and analyses show that our approach is effective: the speaker's utterances become closer to the listener's domain of expertise, which leads to higher communicative success.

Mario Giulianelli, Iris Luden, Raquel Fernandez et al.

We propose using automatically generated natural language definitions of contextualised word usages as interpretable word and word sense representations. Given a collection of usage examples for a target word, and the corresponding data-driven usage clusters (i.e., word senses), a definition is generated for each usage with a specialised Flan-T5 language model, and the most prototypical definition in a usage cluster is chosen as the sense label. We demonstrate how the resulting sense labels can make existing approaches to semantic change analysis more interpretable, and how they can allow users -- historical linguists, lexicographers, or social scientists -- to explore and intuitively explain diachronic trajectories of word meaning. Semantic change analysis is only one of many possible applications of the `definitions as representations' paradigm. Beyond being human-readable, contextualised definitions also outperform token or usage sentence embeddings in word-in-context semantic similarity judgements, making them a new promising type of lexical representation for NLP.

Mario Giulianelli, Andrey Kutuzov, Lidia Pivovarova

Semantics, morphology and syntax are strongly interdependent. However, the majority of computational methods for semantic change detection use distributional word representations which encode mostly semantics. We investigate an alternative method, grammatical profiling, based entirely on changes in the morphosyntactic behaviour of words. We demonstrate that it can be used for semantic change detection and even outperforms some distributional semantic methods. We present an in-depth qualitative and quantitative analysis of the predictions made by our grammatical profiling system, showing that they are plausible and interpretable.

Ece Takmaz, Mario Giulianelli, Sandro Pezzelle et al.

Dialogue participants often refer to entities or situations repeatedly within a conversation, which contributes to its cohesiveness. Subsequent references exploit the common ground accumulated by the interlocutors and hence have several interesting properties, namely, they tend to be shorter and reuse expressions that were effective in previous mentions. In this paper, we tackle the generation of first and subsequent references in visually grounded dialogue. We propose a generation model that produces referring utterances grounded in both the visual and the conversational context. To assess the referring effectiveness of its output, we also implement a reference resolution system. Our experiments and analyses show that the model produces better, more effective referring utterances than a model not grounded in the dialogue context, and generates subsequent references that exhibit linguistic patterns akin to humans.

Andrey Kutuzov, Mario Giulianelli

We apply contextualised word embeddings to lexical semantic change detection in the SemEval-2020 Shared Task 1. This paper focuses on Subtask 2, ranking words by the degree of their semantic drift over time. We analyse the performance of two contextualising architectures (BERT and ELMo) and three change detection algorithms. We find that the most effective algorithms rely on the cosine similarity between averaged token embeddings and the pairwise distances between token embeddings. They outperform strong baselines by a large margin (in the post-evaluation phase, we have the best Subtask 2 submission for SemEval-2020 Task 1), but interestingly, the choice of a particular algorithm depends on the distribution of gold scores in the test set.

Mario Giulianelli, Marco Del Tredici, Raquel Fernández

This paper presents the first unsupervised approach to lexical semantic change that makes use of contextualised word representations. We propose a novel method that exploits the BERT neural language model to obtain representations of word usages, clusters these representations into usage types, and measures change along time with three proposed metrics. We create a new evaluation dataset and show that the model representations and the detected semantic shifts are positively correlated with human judgements. Our extensive qualitative analysis demonstrates that our method captures a variety of synchronic and diachronic linguistic phenomena. We expect our work to inspire further research in this direction.

Mario Giulianelli, Jacqueline Harding, Florian Mohnert et al.

How do neural language models keep track of number agreement between subject and verb? We show that `diagnostic classifiers', trained to predict number from the internal states of a language model, provide a detailed understanding of how, when, and where this information is represented. Moreover, they give us insight into when and where number information is corrupted in cases where the language model ends up making agreement errors. To demonstrate the causal role played by the representations we find, we then use agreement information to influence the course of the LSTM during the processing of difficult sentences. Results from such an intervention reveal a large increase in the language model's accuracy. Together, these results show that diagnostic classifiers give us an unrivalled detailed look into the representation of linguistic information in neural models, and demonstrate that this knowledge can be used to improve their performance.

Mario Giulianelli

There exist two main approaches to automatically extract affective orientation: lexicon-based and corpus-based. In this work, we argue that these two methods are compatible and show that combining them can improve the accuracy of emotion classifiers. In particular, we introduce a novel variant of the Label Propagation algorithm that is tailored to distributed word representations, we apply batch gradient descent to accelerate the optimization of label propagation and to make the optimization feasible for large graphs, and we propose a reproducible method for emotion lexicon expansion. We conclude that label propagation can expand an emotion lexicon in a meaningful way and that the expanded emotion lexicon can be leveraged to improve the accuracy of an emotion classifier.