Luca Malagutti

Clara Meister, Tiago Pimentel, Luca Malagutti et al. · cambridge, harvard

Sampling is a common strategy for generating text from probabilistic models, yet standard ancestral sampling often results in text that is incoherent or ungrammatical. To alleviate this issue, various modifications to a model's sampling distribution, such as nucleus or top-k sampling, have been introduced and are now ubiquitously used in language generation systems. We propose a unified framework for understanding these techniques, which we term sampling adapters. Sampling adapters often lead to qualitatively better text, which raises the question: From a formal perspective, how are they changing the (sub)word-level distributions of language generation models? And why do these local changes lead to higher-quality text? We argue that the shift they enforce can be viewed as a trade-off between precision and recall: while the model loses its ability to produce certain strings, its precision rate on desirable text increases. While this trade-off is not reflected in standard metrics of distribution quality (such as perplexity), we find that several precision-emphasizing measures indeed indicate that sampling adapters can lead to probability distributions more aligned with the true distribution. Further, these measures correlate with higher sequence-level quality scores, specifically, Mauve.

Juan Luis Gastaldi, John Terilla, Luca Malagutti et al.

Tokenization - the practice of converting strings of characters from an alphabet into sequences of tokens over a vocabulary - is a critical step in the NLP pipeline. The use of token representations is widely credited with increased model performance but is also the source of many undesirable behaviors, such as spurious ambiguity or inconsistency. Despite its recognized importance as a standard representation method in NLP, the theoretical underpinnings of tokenization are not yet fully understood. In particular, the impact of tokenization on language model estimation has been investigated primarily through empirical means. The present paper contributes to addressing this theoretical gap by proposing a unified formal framework for representing and analyzing tokenizer models. Based on the category of stochastic maps, this framework enables us to establish general conditions for a principled use of tokenizers and, most importantly, the necessary and sufficient conditions for a tokenizer model to preserve the consistency of statistical estimators. In addition, we discuss statistical and computational concerns crucial for designing and implementing tokenizer models, such as inconsistency, ambiguity, finiteness, and sequentiality. The framework and results advanced in this paper contribute to building robust theoretical foundations for representations in neural language modeling that can inform future theoretical and empirical research.

Luca Malagutti, Andrius Buinovskij, Anej Svete et al. · allen-ai, eth-zurich

For nearly three decades, language models derived from the $n$-gram assumption held the state of the art on the task. The key to their success lay in the application of various smoothing techniques that served to combat overfitting. However, when neural language models toppled $n$-gram models as the best performers, $n$-gram smoothing techniques became less relevant. Indeed, it would hardly be an understatement to suggest that the line of inquiry into $n$-gram smoothing techniques became dormant. This paper re-opens the role classical $n$-gram smoothing techniques may play in the age of neural language models. First, we draw a formal equivalence between label smoothing, a popular regularization technique for neural language models, and add-$λ$ smoothing. Second, we derive a generalized framework for converting any $n$-gram smoothing technique into a regularizer compatible with neural language models. Our empirical results find that our novel regularizers are comparable to and, indeed, sometimes outperform label smoothing on language modeling and machine translation.

Hans Liebl, David Schinz, Anjany Sekuboyina et al.

With the advent of deep learning algorithms, fully automated radiological image analysis is within reach. In spine imaging, several atlas- and shape-based as well as deep learning segmentation algorithms have been proposed, allowing for subsequent automated analysis of morphology and pathology. The first Large Scale Vertebrae Segmentation Challenge (VerSe 2019) showed that these perform well on normal anatomy, but fail in variants not frequently present in the training dataset. Building on that experience, we report on the largely increased VerSe 2020 dataset and results from the second iteration of the VerSe challenge (MICCAI 2020, Lima, Peru). VerSe 2020 comprises annotated spine computed tomography (CT) images from 300 subjects with 4142 fully visualized and annotated vertebrae, collected across multiple centres from four different scanner manufacturers, enriched with cases that exhibit anatomical variants such as enumeration abnormalities (n=77) and transitional vertebrae (n=161). Metadata includes vertebral labelling information, voxel-level segmentation masks obtained with a human-machine hybrid algorithm and anatomical ratings, to enable the development and benchmarking of robust and accurate segmentation algorithms.