Ivan Titov

NeurIPS Conference 2025 Conference Paper

A Controllable Examination for Long-Context Language Models

• Yijun Yang
• Zeyu Huang
• Wenhao Zhu
• Zihan Qiu
• Fei Yuan
• Jeff Pan
• Ivan Titov

Existing frameworks for evaluating long-context language models (LCLM) can be broadly categorized into real-world applications (e. g, document summarization) and synthetic tasks (e. g, needle-in-a-haystack). Despite their utility, both approaches are accompanied by certain intrinsic limitations. Real-world tasks often involve complexity that makes interpretation challenging and suffer from data contamination, whereas synthetic tasks frequently lack meaningful coherence between the target information ("needle") and its surrounding context ("haystack"), undermining their validity as proxies for realistic applications. In response to these challenges, we posit that an ideal long-context evaluation framework should be characterized by three essential features: 1) seamless context: coherent contextual integration between target information and its surrounding context; 2) controllable setting: an extensible task setup that enables controlled studies—for example, incorporating additional required abilities such as numerical reasoning; and 3) sound evaluation: avoiding LLM-as-Judge and conduct exact-match to ensure deterministic and reproducible evaluation results. This study introduces $\textbf{LongBioBench}$, a benchmark that utilizes artificially generated biographies as a controlled environment for assessing LCLMs across dimensions of $\textit{understanding}$, $\textit{reasoning}$, and $\textit{trustworthiness}$. Our experimental evaluation, which includes $\textbf{18}$ LCLMs in total, demonstrates that most models still exhibit deficiencies in semantic understanding and elementary reasoning over retrieved results and are less trustworthy as context length increases. Our further analysis indicates some design choices employed by existing synthetic benchmarks, such as contextual non-coherence, numerical needles, and the absence of distractors, rendering them vulnerable to test the model's long-context capabilities. Moreover, we also reveal that long-context continual pretraining primarily adjusts RoPE embedding to accommodate extended context lengths, which in turn yields only marginal improvements in the model’s true capabilities. To sum up, compared to previous synthetic benchmarks, LongBioBench achieves a better trade-off between mirroring authentic language tasks and maintaining controllability, and is highly interpretable and configurable.

ICLR Conference 2025 Conference Paper

What's New in My Data? Novelty Exploration via Contrastive Generation

• Masaru Isonuma
• Ivan Titov

Fine-tuning is widely used to adapt language models for specific goals, often leveraging real-world data such as patient records, customer-service interactions, or web content in languages not covered in pre-training. These datasets are typically massive, noisy, and often confidential, making their direct inspection challenging. However, understanding them is essential for guiding model deployment and informing decisions about data cleaning or suppressing any harmful behaviors learned during fine-tuning. In this study, we introduce the task of novelty discovery through generation, which aims to identify novel domains of a fine-tuning dataset by generating examples that illustrate these properties. Our approach － Contrastive Generative Exploration (CGE) － assumes no direct access to the data but instead relies on a pre-trained model and the same model after fine-tuning. By contrasting the predictions of these two models, CGE can generate examples that highlight novel domains of the fine-tuning data. However, this simple approach may produce examples that are too similar to one another, failing to capture the full range of novel domains present in the dataset. We address this by introducing an iterative version of CGE, where the previously generated examples are used to update the pre-trained model, and this updated model is then contrasted with the fully fine-tuned model to generate the next example, promoting diversity in the generated outputs. Our experiments demonstrate the effectiveness of CGE in detecting novel domains, such as toxic language, as well as new natural and programming languages. Furthermore, we show that CGE remains effective even when models are fine-tuned using differential privacy techniques.

NeurIPS Conference 2023 Conference Paper

Theoretical and Practical Perspectives on what Influence Functions Do

• Andrea Schioppa
• Katja Filippova
• Ivan Titov
• Polina Zablotskaia

Influence functions (IF) have been seen as a technique for explaining model predictions through the lens of the training data. Their utility is assumed to be in identifying training examples "responsible" for a prediction so that, for example, correcting a prediction is possible by intervening on those examples (removing or editing them) and retraining the model. However, recent empirical studies have shown that the existing methods of estimating IF predict the leave-one-out-and-retrain effect poorly. In order to understand the mismatch between the theoretical promise and the practical results, we analyse five assumptions made by IF methods which are problematic for modern-scale deep neural networks and which concern convexity, numeric stability, training trajectory and parameter divergence. This allows us to clarify what can be expected theoretically from IF. We show that while most assumptions can be addressed successfully, the parameter divergence poses a clear limitation on the predictive power of IF: influence fades over training time even with deterministic training. We illustrate this theoretical result with BERT and ResNet models. Another conclusion from the theoretical analysis is that IF are still useful for model debugging and correcting even though some of the assumptions made in prior work do not hold: using natural language processing and computer vision tasks, we verify that mis-predictions can be successfully corrected by taking only a few fine-tuning steps on influential examples.

NeurIPS Conference 2021 Conference Paper

Structured Reordering for Modeling Latent Alignments in Sequence Transduction

• Bailin Wang
• Mirella Lapata
• Ivan Titov

Despite success in many domains, neural models struggle in settings where train and test examples are drawn from different distributions. In particular, in contrast to humans, conventional sequence-to-sequence (seq2seq) models fail to generalize systematically, i. e. , interpret sentences representing novel combinations of concepts (e. g. , text segments) seen in training. Traditional grammar formalisms excel in such settings by implicitly encoding alignments between input and output segments, but are hard to scale and maintain. Instead of engineering a grammar, we directly model segment-to-segment alignments as discrete structured latent variables within a neural seq2seq model. To efficiently explore the large space of alignments, we introduce a reorder-first align-later framework whose central component is a neural reordering module producing separable permutations. We present an efficient dynamic programming algorithm performing exact marginal inference of separable permutations, and, thus, enabling end-to-end differentiable training of our model. The resulting seq2seq model exhibits better systematic generalization than standard models on synthetic problems and NLP tasks (i. e. , semantic parsing and machine translation).

NeurIPS Conference 2017 Conference Paper

Emergence of Language with Multi-agent Games: Learning to Communicate with Sequences of Symbols

• Serhii Havrylov
• Ivan Titov

Learning to communicate through interaction, rather than relying on explicit supervision, is often considered a prerequisite for developing a general AI. We study a setting where two agents engage in playing a referential game and, from scratch, develop a communication protocol necessary to succeed in this game. Unlike previous work, we require that messages they exchange, both at train and test time, are in the form of a language (i. e. sequences of discrete symbols). We compare a reinforcement learning approach and one using a differentiable relaxation (straight-through Gumbel-softmax estimator) and observe that the latter is much faster to converge and it results in more effective protocols. Interestingly, we also observe that the protocol we induce by optimizing the communication success exhibits a degree of compositionality and variability (i. e. the same information can be phrased in different ways), both properties characteristic of natural languages. As the ultimate goal is to ensure that communication is accomplished in natural language, we also perform experiments where we inject prior information about natural language into our model and study properties of the resulting protocol.

JMLR Journal 2010 Journal Article

Incremental Sigmoid Belief Networks for Grammar Learning

• James Henderson
• Ivan Titov

We propose a class of Bayesian networks appropriate for structured prediction problems where the Bayesian network's model structure is a function of the predicted output structure. These incremental sigmoid belief networks (ISBNs) make decoding possible because inference with partial output structures does not require summing over the unboundedly many compatible model structures, due to their directed edges and incrementally specified model structure. ISBNs are specifically targeted at challenging structured prediction problems such as natural language parsing, where learning the domain's complex statistical dependencies benefits from large numbers of latent variables. While exact inference in ISBNs with large numbers of latent variables is not tractable, we propose two efficient approximations. First, we demonstrate that a previous neural network parsing model can be viewed as a coarse mean-field approximation to inference with ISBNs. We then derive a more accurate but still tractable variational approximation, which proves effective in artificial experiments. We compare the effectiveness of these models on a benchmark natural language parsing task, where they achieve accuracy competitive with the state-of-the-art. The model which is a closer approximation to an ISBN has better parsing accuracy, suggesting that ISBNs are an appropriate abstract model of natural language grammar learning. [abs] [ pdf ][ bib ] &copy JMLR 2010. ( edit, beta )

IJCAI Conference 2009 Conference Paper

• Ivan Titov
• James Henderson
• Paola Merlo
• Gabriele Musillo

This paper investigates a generative history-based parsing model that synchronises the derivation of non-planar graphs representing semantic dependencies with the derivation of dependency trees representing syntactic structures. To process non-planarity online, the semantic transition-based parser uses a new technique to dynamically reorder nodes during the derivation. While the synchronised derivations allow different structures to be built for the semantic non-planar graphs and syntactic dependency trees, useful statistical dependencies between these structures are modeled using latent variables. The resulting synchronous parser achieves competitive performance on the CoNLL- 2008 shared task, achieving relative error reduction of 12% in semantic F score over previously proposed synchronous models that cannot process non-planarity online.

IJCAI Conference 2009 Conference Paper

• Alexandre Klementiev
• Dan Roth
• Kevin Small
• Ivan Titov

Consider the setting where a panel of judges is repeatedly asked to (partially) rank sets of objects according to given criteria, and assume that the judges’ expertise depends on the objects’ domain. Learning to aggregate their rankings with the goal of producing a better joint ranking is a fundamental problem in many areas of Information Retrieval and Natural Language Processing, amongst others. However, supervised ranking data is generally dif- ficult to obtain, especially if coming from multiple domains. Therefore, we propose a framework for learning to aggregate votes of constituent rankers with domain specific expertise without supervision. We apply the learning framework to the settings of aggregating full rankings and aggregating top-k lists, demonstrating significant improvements over a domain-agnostic baseline in both cases.