James Henderson

Rabeeh Karimi Mahabadi, Luke Zettlemoyer, James Henderson et al. · meta-ai, uw

Current methods for few-shot fine-tuning of pretrained masked language models (PLMs) require carefully engineered prompts and verbalizers for each new task to convert examples into a cloze-format that the PLM can score. In this work, we propose PERFECT, a simple and efficient method for few-shot fine-tuning of PLMs without relying on any such handcrafting, which is highly effective given as few as 32 data points. PERFECT makes two key design choices: First, we show that manually engineered task prompts can be replaced with task-specific adapters that enable sample-efficient fine-tuning and reduce memory and storage costs by roughly factors of 5 and 100, respectively. Second, instead of using handcrafted verbalizers, we learn new multi-token label embeddings during fine-tuning, which are not tied to the model vocabulary and which allow us to avoid complex auto-regressive decoding. These embeddings are not only learnable from limited data but also enable nearly 100x faster training and inference. Experiments on a wide range of few-shot NLP tasks demonstrate that PERFECT, while being simple and efficient, also outperforms existing state-of-the-art few-shot learning methods. Our code is publicly available at https://github.com/facebookresearch/perfect.git.

Alireza Mohammadshahi, Vassilina Nikoulina, Alexandre Berard et al.

In recent years, multilingual machine translation models have achieved promising performance on low-resource language pairs by sharing information between similar languages, thus enabling zero-shot translation. To overcome the "curse of multilinguality", these models often opt for scaling up the number of parameters, which makes their use in resource-constrained environments challenging. We introduce SMaLL-100, a distilled version of the M2M-100 (12B) model, a massively multilingual machine translation model covering 100 languages. We train SMaLL-100 with uniform sampling across all language pairs and therefore focus on preserving the performance of low-resource languages. We evaluate SMaLL-100 on different low-resource benchmarks: FLORES-101, Tatoeba, and TICO-19 and demonstrate that it outperforms previous massively multilingual models of comparable sizes (200-600M) while improving inference latency and memory usage. Additionally, our model achieves comparable results to M2M-100 (1.2B), while being 3.6x smaller and 4.3x faster at inference. Code and pre-trained models: https://github.com/alirezamshi/small100

Alireza Mohammadshahi, Thomas Scialom, Majid Yazdani et al. · meta-ai

Existing metrics for evaluating the quality of automatically generated questions such as BLEU, ROUGE, BERTScore, and BLEURT compare the reference and predicted questions, providing a high score when there is a considerable lexical overlap or semantic similarity between the candidate and the reference questions. This approach has two major shortcomings. First, we need expensive human-provided reference questions. Second, it penalises valid questions that may not have high lexical or semantic similarity to the reference questions. In this paper, we propose a new metric, RQUGE, based on the answerability of the candidate question given the context. The metric consists of a question-answering and a span scorer modules, using pre-trained models from existing literature, thus it can be used without any further training. We demonstrate that RQUGE has a higher correlation with human judgment without relying on the reference question. Additionally, RQUGE is shown to be more robust to several adversarial corruptions. Furthermore, we illustrate that we can significantly improve the performance of QA models on out-of-domain datasets by fine-tuning on synthetic data generated by a question generation model and re-ranked by RQUGE.

Alireza Mohammadshahi, Vassilina Nikoulina, Alexandre Berard et al.

Recently, very large pre-trained models achieve state-of-the-art results in various natural language processing (NLP) tasks, but their size makes it more challenging to apply them in resource-constrained environments. Compression techniques allow to drastically reduce the size of the models and therefore their inference time with negligible impact on top-tier metrics. However, the general performance averaged across multiple tasks and/or languages may hide a drastic performance drop on under-represented features, which could result in the amplification of biases encoded by the models. In this work, we assess the impact of compression methods on Multilingual Neural Machine Translation models (MNMT) for various language groups, gender, and semantic biases by extensive analysis of compressed models on different machine translation benchmarks, i.e. FLORES-101, MT-Gender, and DiBiMT. We show that the performance of under-represented languages drops significantly, while the average BLEU metric only slightly decreases. Interestingly, the removal of noisy memorization with compression leads to a significant improvement for some medium-resource languages. Finally, we demonstrate that compression amplifies intrinsic gender and semantic biases, even in high-resource languages. Code: https://github.com/alirezamshi/bias-compressedMT

Florian Mai, Arnaud Pannatier, Fabio Fehr et al.

Transformer-based architectures are the model of choice for natural language understanding, but they come at a significant cost, as they have quadratic complexity in the input length, require a lot of training data, and can be difficult to tune. In the pursuit of lower costs, we investigate simple MLP-based architectures. We find that existing architectures such as MLPMixer, which achieves token mixing through a static MLP applied to each feature independently, are too detached from the inductive biases required for natural language understanding. In this paper, we propose a simple variant, HyperMixer, which forms the token mixing MLP dynamically using hypernetworks. Empirically, we demonstrate that our model performs better than alternative MLP-based models, and on par with Transformers. In contrast to Transformers, HyperMixer achieves these results at substantially lower costs in terms of processing time, training data, and hyperparameter tuning.

Christos Theodoropoulos, Andrei Catalin Coman, James Henderson et al.

The ever-growing volume of biomedical publications creates a critical need for efficient knowledge discovery. In this context, we introduce an open-source end-to-end framework designed to construct knowledge around specific diseases directly from raw text. To facilitate research in disease-related knowledge discovery, we create two annotated datasets focused on Rett syndrome and Alzheimer's disease, enabling the identification of semantic relations between biomedical entities. Extensive benchmarking explores various ways to represent relations and entity representations, offering insights into optimal modeling strategies for semantic relation detection and highlighting language models' competence in knowledge discovery. We also conduct probing experiments using different layer representations and attention scores to explore transformers' ability to capture semantic relations.

James Henderson, Alireza Mohammadshahi, Andrei C. Coman et al.

We argue that Transformers are essentially graph-to-graph models, with sequences just being a special case. Attention weights are functionally equivalent to graph edges. Our Graph-to-Graph Transformer architecture makes this ability explicit, by inputting graph edges into the attention weight computations and predicting graph edges with attention-like functions, thereby integrating explicit graphs into the latent graphs learned by pretrained Transformers. Adding iterative graph refinement provides a joint embedding of input, output, and latent graphs, allowing non-autoregressive graph prediction to optimise the complete graph without any bespoke pipeline or decoding strategy. Empirical results show that this architecture achieves state-of-the-art accuracies for modelling a variety of linguistic structures, integrating very effectively with the latent linguistic representations learned by pretraining.

Melika Behjati, Fabio Fehr, James Henderson

Learned representations at the level of characters, sub-words, words and sentences, have each contributed to advances in understanding different NLP tasks and linguistic phenomena. However, learning textual embeddings is costly as they are tokenization specific and require different models to be trained for each level of abstraction. We introduce a novel language representation model which can learn to compress to different levels of abstraction at different layers of the same model. We apply Nonparametric Variational Information Bottleneck (NVIB) to stacked Transformer self-attention layers in the encoder, which encourages an information-theoretic compression of the representations through the model. We find that the layers within the model correspond to increasing levels of abstraction and that their representations are more linguistically informed. Finally, we show that NVIB compression results in a model which is more robust to adversarial perturbations.

Luis Espinosa-Anke, Alexander Shvets, Alireza Mohammadshahi et al.

Recognizing and categorizing lexical collocations in context is useful for language learning, dictionary compilation and downstream NLP. However, it is a challenging task due to the varying degrees of frozenness lexical collocations exhibit. In this paper, we put forward a sequence tagging BERT-based model enhanced with a graph-aware transformer architecture, which we evaluate on the task of collocation recognition in context. Our results suggest that explicitly encoding syntactic dependencies in the model architecture is helpful, and provide insights on differences in collocation typification in English, Spanish and French.

Andrei C. Coman, Christos Theodoropoulos, Marie-Francine Moens et al.

Document-level relation extraction typically relies on text-based encoders and hand-coded pooling heuristics to aggregate information learned by the encoder. In this paper, we leverage the intrinsic graph processing capabilities of the Transformer model and propose replacing hand-coded pooling methods with new tokens in the input, which are designed to aggregate information via explicit graph relations in the computation of attention weights. We introduce a joint text-graph Transformer model and a graph-assisted declarative pooling (GADePo) specification of the input, which provides explicit and high-level instructions for information aggregation. GADePo allows the pooling process to be guided by domain-specific knowledge or desired outcomes but still learned by the Transformer, leading to more flexible and customisable pooling strategies. We evaluate our method across diverse datasets and models and show that our approach yields promising results that are consistently better than those achieved by the hand-coded pooling functions.

James Henderson, Fabio Fehr

We propose a VAE for Transformers by developing a variational information bottleneck regulariser for Transformer embeddings. We formalise the embedding space of Transformer encoders as mixture probability distributions, and use Bayesian nonparametrics to derive a nonparametric variational information bottleneck (NVIB) for such attention-based embeddings. The variable number of mixture components supported by nonparametric methods captures the variable number of vectors supported by attention, and the exchangeability of our nonparametric distributions captures the permutation invariance of attention. This allows NVIB to regularise the number of vectors accessible with attention, as well as the amount of information in individual vectors. By regularising the cross-attention of a Transformer encoder-decoder with NVIB, we propose a nonparametric variational autoencoder (NVAE). Initial experiments on training a NVAE on natural language text show that the induced embedding space has the desired properties of a VAE for Transformers.

Andrei C. Coman, Christos Theodoropoulos, Marie-Francine Moens et al.

We propose Fast-and-Frugal Text-Graph (FnF-TG) Transformers, a Transformer-based framework that unifies textual and structural information for inductive link prediction in text-attributed knowledge graphs. We demonstrate that, by effectively encoding ego-graphs (1-hop neighbourhoods), we can reduce the reliance on resource-intensive textual encoders. This makes the model both fast at training and inference time, as well as frugal in terms of cost. We perform a comprehensive evaluation on three popular datasets and show that FnF-TG can achieve superior performance compared to previous state-of-the-art methods. We also extend inductive learning to a fully inductive setting, where relations don't rely on transductive (fixed) representations, as in previous work, but are a function of their textual description. Additionally, we introduce new variants of existing datasets, specifically designed to test the performance of models on unseen relations at inference time, thus offering a new test-bench for fully inductive link prediction.

Rabeeh Karimi Mahabadi, Hamish Ivison, Jaesung Tae et al.

Diffusion models have emerged as a powerful paradigm for generation, obtaining strong performance in various continuous domains. However, applying continuous diffusion models to natural language remains challenging due to its discrete nature and the need for a large number of diffusion steps to generate text, making diffusion-based generation expensive. In this work, we propose Text-to-text Self-conditioned Simplex Diffusion (TESS), a text diffusion model that is fully non-autoregressive, employs a new form of self-conditioning, and applies the diffusion process on the logit simplex space rather than the learned embedding space. Through extensive experiments on natural language understanding and generation tasks including summarization, text simplification, paraphrase generation, and question generation, we demonstrate that TESS outperforms state-of-the-art non-autoregressive models, requires fewer diffusion steps with minimal drop in performance, and is competitive with pretrained autoregressive sequence-to-sequence models. We publicly release our codebase at https://github.com/allenai/tess-diffusion.

Rabeeh Karimi Mahabadi, Yonatan Belinkov, James Henderson

While large-scale pretrained language models have obtained impressive results when fine-tuned on a wide variety of tasks, they still often suffer from overfitting in low-resource scenarios. Since such models are general-purpose feature extractors, many of these features are inevitably irrelevant for a given target task. We propose to use Variational Information Bottleneck (VIB) to suppress irrelevant features when fine-tuning on low-resource target tasks, and show that our method successfully reduces overfitting. Moreover, we show that our VIB model finds sentence representations that are more robust to biases in natural language inference datasets, and thereby obtains better generalization to out-of-domain datasets. Evaluation on seven low-resource datasets in different tasks shows that our method significantly improves transfer learning in low-resource scenarios, surpassing prior work. Moreover, it improves generalization on 13 out of 15 out-of-domain natural language inference benchmarks. Our code is publicly available in https://github.com/rabeehk/vibert.

Rabeeh Karimi Mahabadi, James Henderson, Sebastian Ruder

Adapting large-scale pretrained language models to downstream tasks via fine-tuning is the standard method for achieving state-of-the-art performance on NLP benchmarks. However, fine-tuning all weights of models with millions or billions of parameters is sample-inefficient, unstable in low-resource settings, and wasteful as it requires storing a separate copy of the model for each task. Recent work has developed parameter-efficient fine-tuning methods, but these approaches either still require a relatively large number of parameters or underperform standard fine-tuning. In this work, we propose Compacter, a method for fine-tuning large-scale language models with a better trade-off between task performance and the number of trainable parameters than prior work. Compacter accomplishes this by building on top of ideas from adapters, low-rank optimization, and parameterized hypercomplex multiplication layers. Specifically, Compacter inserts task-specific weight matrices into a pretrained model's weights, which are computed efficiently as a sum of Kronecker products between shared "slow" weights and "fast" rank-one matrices defined per Compacter layer. By only training 0.047% of a pretrained model's parameters, Compacter performs on par with standard fine-tuning on GLUE and outperforms standard fine-tuning on SuperGLUE and low-resource settings. Our code is publicly available at~\url{https://github.com/rabeehk/compacter}.

Rabeeh Karimi Mahabadi, Sebastian Ruder, Mostafa Dehghani et al.

State-of-the-art parameter-efficient fine-tuning methods rely on introducing adapter modules between the layers of a pretrained language model. However, such modules are trained separately for each task and thus do not enable sharing information across tasks. In this paper, we show that we can learn adapter parameters for all layers and tasks by generating them using shared hypernetworks, which condition on task, adapter position, and layer id in a transformer model. This parameter-efficient multi-task learning framework allows us to achieve the best of both worlds by sharing knowledge across tasks via hypernetworks while enabling the model to adapt to each individual task through task-specific adapters. Experiments on the well-known GLUE benchmark show improved performance in multi-task learning while adding only 0.29% parameters per task. We additionally demonstrate substantial performance improvements in few-shot domain generalization across a variety of tasks. Our code is publicly available in https://github.com/rabeehk/hyperformer.

Rabeeh Karimi Mahabadi, Yonatan Belinkov, James Henderson

Several recent studies have shown that strong natural language understanding (NLU) models are prone to relying on unwanted dataset biases without learning the underlying task, resulting in models that fail to generalize to out-of-domain datasets and are likely to perform poorly in real-world scenarios. We propose two learning strategies to train neural models, which are more robust to such biases and transfer better to out-of-domain datasets. The biases are specified in terms of one or more bias-only models, which learn to leverage the dataset biases. During training, the bias-only models' predictions are used to adjust the loss of the base model to reduce its reliance on biases by down-weighting the biased examples and focusing the training on the hard examples. We experiment on large-scale natural language inference and fact verification benchmarks, evaluating on out-of-domain datasets that are specifically designed to assess the robustness of models against known biases in the training data. Results show that our debiasing methods greatly improve robustness in all settings and better transfer to other textual entailment datasets. Our code and data are publicly available in \url{https://github.com/rabeehk/robust-nli}.

Dina El Zein, Shashi Kumar, James Henderson

The nonparametric variational information bottleneck (NVIB) provides the foundation for nonparametric variational differential privacy (NVDP), a framework for building privacy-preserving language models. However, the learned latent representations can drift into regions with high information content, leading to poor privacy guarantees, but also low utility due to numerical instability during training. In this work, we introduce a principled parameter clipping strategy to directly address this issue. Our method is mathematically derived from the objective of minimizing the Rényi Divergence (RD) upper bound, yielding specific, theoretically grounded constraints on the posterior mean, variance, and mixture weight parameters. We apply our technique to an NVIB based model and empirically compare it against an unconstrained baseline. Our findings demonstrate that the clipped model consistently achieves tighter RD bounds, implying stronger privacy, while simultaneously attaining higher performance on several downstream tasks. This work presents a simple yet effective method for improving the privacy-utility trade-off in variational models, making them more robust and practical.

Melika Behjati, James Henderson

Fine-grained knowledge is crucial for vision-language models to obtain a better understanding of the real world. While there has been work trying to acquire this kind of knowledge in the space of vision and language, it has mostly focused on aligning the image patches with the tokens on the language side. However, image patches do not have any meaning to the human eye, and individual tokens do not necessarily carry groundable information in the image. It is groups of tokens which describe different aspects of the scene. In this work, we propose a model which groups the caption tokens as part of its architecture in order to capture a fine-grained representation of the language. We expect our representations to be at the level of objects present in the image, and therefore align our representations with the output of an image encoder trained to discover objects. We show that by learning to group the tokens, the vision-language model has a better fine-grained understanding of vision and language. In addition, the token groups that our model discovers are highly similar to groundable phrases in text, both qualitatively and quantitatively.

Dina El Zein, James Henderson

We propose a privacy-preserving method for sharing text data by sharing noisy versions of their transformer embeddings. It has been shown that hidden representations learned by deep models can encode sensitive information from the input, making it possible for adversaries to recover the input data with considerable accuracy. This problem is exacerbated in transformer embeddings because they consist of multiple vectors, one per token. To mitigate this risk, we propose Nonparametric Variational Differential Privacy (NVDP), which ensures both useful data sharing and strong privacy protection. We take a differential privacy (DP) approach, integrating a nonparametric variational information bottleneck (NVIB) layer into the transformer architecture to inject noise into its multivector embeddings and thereby hide information, and measuring privacy protection with Rényi Divergence (RD) and its corresponding Bayesian Differential Privacy (BDP) guarantee. Training the NVIB layer calibrates the noise level according to the utility of the downstream task. We test NVDP on the General Language Understanding Evaluation (GLUE) benchmark and show that varying the noise level gives us a useful trade-off between privacy and accuracy. With lower noise levels, our model maintains high accuracy while offering strong privacy guarantees, effectively balancing privacy and utility.

Andrei C. Coman, Ionut-Teodor Sorodoc, Leonardo F. R. Ribeiro et al.

Existing Reward Models (RMs), typically trained on general preference data, struggle in Retrieval Augmented Generation (RAG) settings, which require judging responses for faithfulness to retrieved context, relevance to the user query, appropriate refusals when context is insufficient, completeness and conciseness of information. To address the lack of publicly available RAG-centric preference datasets and specialised RMs, we introduce RAGferee, a methodology that repurposes question-answering (QA) datasets into preference pairs that prioritise groundedness over stylistic features, enabling the training of contextual RMs better suited to judging RAG responses. Using RAGferee, we curate a small preference dataset of 4K samples and fine-tune RMs ranging from 7B to 24B parameters. Our RAG-centric RMs achieve state-of-the-art performance on ContextualJudgeBench, surpassing existing 70B+ RMs trained on much larger (up to 2.4M samples) general corpora, with an absolute improvement of +15.5%.

Christos Theodoropoulos, Andrei Catalin Coman, James Henderson et al.

Knowledge discovery is hindered by the increasing volume of publications and the scarcity of extensive annotated data. To tackle the challenge of information overload, it is essential to employ automated methods for knowledge extraction and processing. Finding the right balance between the level of supervision and the effectiveness of models poses a significant challenge. While supervised techniques generally result in better performance, they have the major drawback of demanding labeled data. This requirement is labor-intensive and time-consuming and hinders scalability when exploring new domains. In this context, our study addresses the challenge of identifying semantic relationships between biomedical entities (e.g., diseases, proteins) in unstructured text while minimizing dependency on supervision. We introduce a suite of unsupervised algorithms based on dependency trees and attention mechanisms and employ a range of pointwise binary classification methods. Transitioning from weakly supervised to fully unsupervised settings, we assess the methods' ability to learn from data with noisy labels. The evaluation on biomedical benchmark datasets explores the effectiveness of the methods. Our approach tackles a central issue in knowledge discovery: balancing performance with minimal supervision. By gradually decreasing supervision, we assess the robustness of pointwise binary classification techniques in handling noisy labels, revealing their capability to shift from weakly supervised to entirely unsupervised scenarios. Comprehensive benchmarking offers insights into the effectiveness of these techniques, suggesting an encouraging direction toward adaptable knowledge discovery systems, representing progress in creating data-efficient methodologies for extracting useful insights when annotated data is limited.

Andrew G. T. Pyo, Yuta Nagano, Martina Milighetti et al.

The biophysical interactions between the T cell receptor (TCR) and its ligands determine the specificity of the cellular immune response. However, the immense diversity of receptors and ligands has made it challenging to discover generalizable rules across the distinct binding affinity landscapes created by different ligands. Here, we present an optimization framework for discovering biophysical rules that predict whether TCRs share specificity to a ligand. Applying this framework to TCRs associated with a collection of SARS-CoV-2 peptides we systematically characterize how co-specificity depends on the type and position of amino-acid differences between receptors. We also demonstrate that the inferred rules generalize to ligands highly dissimilar to any seen during training. Our analysis reveals that matching of steric properties between substituted amino acids is more important for receptor co-specificity than the hydrophobic properties that prominently determine evolutionary substitutability. Our analysis also quantifies the substantial importance of positions not in direct contact with the peptide for specificity. These findings highlight the potential for data-driven approaches to uncover the molecular mechanisms underpinning the specificity of adaptive immune responses.

Yuta Nagano, Andrew Pyo, Martina Milighetti et al.

Computational prediction of the interaction of T cell receptors (TCRs) and their ligands is a grand challenge in immunology. Despite advances in high-throughput assays, specificity-labelled TCR data remains sparse. In other domains, the pre-training of language models on unlabelled data has been successfully used to address data bottlenecks. However, it is unclear how to best pre-train protein language models for TCR specificity prediction. Here we introduce a TCR language model called SCEPTR (Simple Contrastive Embedding of the Primary sequence of T cell Receptors), capable of data-efficient transfer learning. Through our model, we introduce a novel pre-training strategy combining autocontrastive learning and masked-language modelling, which enables SCEPTR to achieve its state-of-the-art performance. In contrast, existing protein language models and a variant of SCEPTR pre-trained without autocontrastive learning are outperformed by sequence alignment-based methods. We anticipate that contrastive learning will be a useful paradigm to decode the rules of TCR specificity.

Lesly Miculicich, James Henderson

The state-of-the-art models for coreference resolution are based on independent mention pair-wise decisions. We propose a modelling approach that learns coreference at the document-level and takes global decisions. For this purpose, we model coreference links in a graph structure where the nodes are tokens in the text, and the edges represent the relationship between them. Our model predicts the graph in a non-autoregressive manner, then iteratively refines it based on previous predictions, allowing global dependencies between decisions. The experimental results show improvements over various baselines, reinforcing the hypothesis that document-level information improves conference resolution.

Florian Mai, James Henderson

Text autoencoders are often used for unsupervised conditional text generation by applying mappings in the latent space to change attributes to the desired values. Recently, Mai et al. (2020) proposed Emb2Emb, a method to learn these mappings in the embedding space of an autoencoder. However, their method is restricted to autoencoders with a single-vector embedding, which limits how much information can be retained. We address this issue by extending their method to Bag-of-Vectors Autoencoders (BoV-AEs), which encode the text into a variable-size bag of vectors that grows with the size of the text, as in attention-based models. This allows to encode and reconstruct much longer texts than standard autoencoders. Analogous to conventional autoencoders, we propose regularization techniques that facilitate learning meaningful operations in the latent space. Finally, we adapt Emb2Emb for a training scheme that learns to map an input bag to an output bag, including a novel loss function and neural architecture. Our empirical evaluations on unsupervised sentiment transfer show that our method performs substantially better than a standard autoencoder.

Christos Theodoropoulos, James Henderson, Andrei C. Coman et al.

Though language model text embeddings have revolutionized NLP research, their ability to capture high-level semantic information, such as relations between entities in text, is limited. In this paper, we propose a novel contrastive learning framework that trains sentence embeddings to encode the relations in a graph structure. Given a sentence (unstructured text) and its graph, we use contrastive learning to impose relation-related structure on the token-level representations of the sentence obtained with a CharacterBERT (El Boukkouri et al.,2020) model. The resulting relation-aware sentence embeddings achieve state-of-the-art results on the relation extraction task using only a simple KNN classifier, thereby demonstrating the success of the proposed method. Additional visualization by a tSNE analysis shows the effectiveness of the learned representation space compared to baselines. Furthermore, we show that we can learn a different space for named entity recognition, again using a contrastive learning objective, and demonstrate how to successfully combine both representation spaces in an entity-relation task.

James Barry, Alireza Mohammadshahi, Joachim Wagner et al.

We describe the DCU-EPFL submission to the IWPT 2021 Shared Task on Parsing into Enhanced Universal Dependencies. The task involves parsing Enhanced UD graphs, which are an extension of the basic dependency trees designed to be more facilitative towards representing semantic structure. Evaluation is carried out on 29 treebanks in 17 languages and participants are required to parse the data from each language starting from raw strings. Our approach uses the Stanza pipeline to preprocess the text files, XLMRoBERTa to obtain contextualized token representations, and an edge-scoring and labeling model to predict the enhanced graph. Finally, we run a post-processing script to ensure all of our outputs are valid Enhanced UD graphs. Our system places 6th out of 9 participants with a coarse Enhanced Labeled Attachment Score (ELAS) of 83.57. We carry out additional post-deadline experiments which include using Trankit for pre-processing, XLM-RoBERTa-LARGE, treebank concatenation, and multitask learning between a basic and an enhanced dependency parser. All of these modifications improve our initial score and our final system has a coarse ELAS of 88.04.

Chris McMahon Stone, Sam L. Thomas, Mathy Vanhoef et al.

In this paper, we propose a new approach to infer state machine models from protocol implementations. Our method, STATEINSPECTOR, learns protocol states by using novel program analyses to combine observations of run-time memory and I/O. It requires no access to source code and only lightweight execution monitoring of the implementation under test. We demonstrate and evaluate STATEINSPECTOR's effectiveness on numerous TLS and WPA/2 implementations. In the process, we show STATEINSPECTOR enables deeper state discovery, increased learning efficiency, and more insightful post-mortem analyses than existing approaches. Further to improved learning, our method led us to discover several concerning deviations from the standards and a high impact vulnerability in a prominent Wi-Fi implementation.

Alireza Mohammadshahi, James Henderson

Recent models have shown that incorporating syntactic knowledge into the semantic role labelling (SRL) task leads to a significant improvement. In this paper, we propose Syntax-aware Graph-to-Graph Transformer (SynG2G-Tr) model, which encodes the syntactic structure using a novel way to input graph relations as embeddings, directly into the self-attention mechanism of Transformer. This approach adds a soft bias towards attention patterns that follow the syntactic structure but also allows the model to use this information to learn alternative patterns. We evaluate our model on both span-based and dependency-based SRL datasets, and outperform previous alternative methods in both in-domain and out-of-domain settings, on CoNLL 2005 and CoNLL 2009 datasets.

Melika Behjati, James Henderson

Characters do not convey meaning, but sequences of characters do. We propose an unsupervised distributional method to learn the abstract meaningful units in a sequence of characters. Rather than segmenting the sequence, our Dynamic Capacity Slot Attention model discovers continuous representations of the objects in the sequence, extending an architecture for object discovery in images. We train our model on different languages and evaluate the quality of the obtained representations with forward and reverse probing classifiers. These experiments show that our model succeeds in discovering units which are similar to those proposed previously in form, content and level of abstraction, and which show promise for capturing meaningful information at a higher level of abstraction.

Haozhou Wang, James Henderson, Paola Merlo

Generative adversarial networks (GANs) have succeeded in inducing cross-lingual word embeddings -- maps of matching words across languages -- without supervision. Despite these successes, GANs' performance for the difficult case of distant languages is still not satisfactory. These limitations have been explained by GANs' incorrect assumption that source and target embedding spaces are related by a single linear mapping and are approximately isomorphic. We assume instead that, especially across distant languages, the mapping is only piece-wise linear, and propose a multi-adversarial learning method. This novel method induces the seed cross-lingual dictionary through multiple mappings, each induced to fit the mapping for one subspace. Our experiments on unsupervised bilingual lexicon induction show that this method improves performance over previous single-mapping methods, especially for distant languages.

Florian Mai, Nikolaos Pappas, Ivan Montero et al.

Text autoencoders are commonly used for conditional generation tasks such as style transfer. We propose methods which are plug and play, where any pretrained autoencoder can be used, and only require learning a mapping within the autoencoder's embedding space, training embedding-to-embedding (Emb2Emb). This reduces the need for labeled training data for the task and makes the training procedure more efficient. Crucial to the success of this method is a loss term for keeping the mapped embedding on the manifold of the autoencoder and a mapping which is trained to navigate the manifold by learning offset vectors. Evaluations on style transfer tasks both with and without sequence-to-sequence supervision show that our method performs better than or comparable to strong baselines while being up to four times faster.

James Henderson

In this paper, we trace the history of neural networks applied to natural language understanding tasks, and identify key contributions which the nature of language has made to the development of neural network architectures. We focus on the importance of variable binding and its instantiation in attention-based models, and argue that Transformer is not a sequence model but an induced-structure model. This perspective leads to predictions of the challenges facing research in deep learning architectures for natural language understanding.

Alireza Mohammadshahi, James Henderson

We propose the Recursive Non-autoregressive Graph-to-Graph Transformer architecture (RNGTr) for the iterative refinement of arbitrary graphs through the recursive application of a non-autoregressive Graph-to-Graph Transformer and apply it to syntactic dependency parsing. We demonstrate the power and effectiveness of RNGTr on several dependency corpora, using a refinement model pre-trained with BERT. We also introduce Syntactic Transformer (SynTr), a non-recursive parser similar to our refinement model. RNGTr can improve the accuracy of a variety of initial parsers on 13 languages from the Universal Dependencies Treebanks, English and Chinese Penn Treebanks, and the German CoNLL2009 corpus, even improving over the new state-of-the-art results achieved by SynTr, significantly improving the state-of-the-art for all corpora tested.

Alireza Mohammadshahi, James Henderson

We propose the Graph2Graph Transformer architecture for conditioning on and predicting arbitrary graphs, and apply it to the challenging task of transition-based dependency parsing. After proposing two novel Transformer models of transition-based dependency parsing as strong baselines, we show that adding the proposed mechanisms for conditioning on and predicting graphs of Graph2Graph Transformer results in significant improvements, both with and without BERT pre-training. The novel baselines and their integration with Graph2Graph Transformer significantly outperform the state-of-the-art in traditional transition-based dependency parsing on both English Penn Treebank, and 13 languages of Universal Dependencies Treebanks. Graph2Graph Transformer can be integrated with many previous structured prediction methods, making it easy to apply to a wide range of NLP tasks.

Lesly Miculicich, James Henderson

Learning to detect entity mentions without using syntactic information can be useful for integration and joint optimization with other tasks. However, it is common to have partially annotated data for this problem. Here, we investigate two approaches to deal with partial annotation of mentions: weighted loss and soft-target classification. We also propose two neural mention detection approaches: a sequence tagging, and an exhaustive search. We evaluate our methods with coreference resolution as a downstream task, using multitask learning. The results show that the recall and F1 score improve for all methods.

Navid Rekabsaz, Nikolaos Pappas, James Henderson et al.

Neural language modeling (LM) has led to significant improvements in several applications, including Automatic Speech Recognition. However, they typically require large amounts of training data, which is not available for many domains and languages. In this study, we propose a multilingual neural language model architecture, trained jointly on the domain-specific data of several low-resource languages. The proposed multilingual LM consists of language specific word embeddings in the encoder and decoder, and one language specific LSTM layer, plus two LSTM layers with shared parameters across the languages. This multilingual LM model facilitates transfer learning across the languages, acting as an extra regularizer in very low-resource scenarios. We integrate our proposed multilingual approach with a state-of-the-art highly-regularized neural LM, and evaluate on the conversational data domain for four languages over a range of training data sizes. Compared to monolingual LMs, the results show significant improvements of our proposed multilingual LM when the amount of available training data is limited, indicating the advantages of cross-lingual parameter sharing in very low-resource language modeling.

Nikolaos Pappas, James Henderson

Many tasks, including language generation, benefit from learning the structure of the output space, particularly when the space of output labels is large and the data is sparse. State-of-the-art neural language models indirectly capture the output space structure in their classifier weights since they lack parameter sharing across output labels. Learning shared output label mappings helps, but existing methods have limited expressivity and are prone to overfitting. In this paper, we investigate the usefulness of more powerful shared mappings for output labels, and propose a deep residual output mapping with dropout between layers to better capture the structure of the output space and avoid overfitting. Evaluations on three language generation tasks show that our output label mapping can match or improve state-of-the-art recurrent and self-attention architectures, and suggest that the classifier does not necessarily need to be high-rank to better model natural language if it is better at capturing the structure of the output space.

Haozhou Wang, James Henderson, Paola Merlo

Distributed representations of words which map each word to a continuous vector have proven useful in capturing important linguistic information not only in a single language but also across different languages. Current unsupervised adversarial approaches show that it is possible to build a mapping matrix that align two sets of monolingual word embeddings together without high quality parallel data such as a dictionary or a sentence-aligned corpus. However, without post refinement, the performance of these methods' preliminary mapping is not good, leading to poor performance for typologically distant languages. In this paper, we propose a weakly-supervised adversarial training method to overcome this limitation, based on the intuition that mapping across languages is better done at the concept level than at the word level. We propose a concept-based adversarial training method which for most languages improves the performance of previous unsupervised adversarial methods, especially for typologically distant language pairs.

Navid Rekabsaz, Robert West, James Henderson et al.

Text corpora are widely used resources for measuring societal biases and stereotypes. The common approach to measuring such biases using a corpus is by calculating the similarities between the embedding vector of a word (like nurse) and the vectors of the representative words of the concepts of interest (such as genders). In this study, we show that, depending on what one aims to quantify as bias, this commonly-used approach can introduce non-relevant concepts into bias measurement. We propose an alternative approach to bias measurement utilizing the smoothed first-order co-occurrence relations between the word and the representative concept words, which we derive by reconstructing the co-occurrence estimates inherent in word embedding models. We compare these approaches by conducting several experiments on the scenario of measuring gender bias of occupational words, according to an English Wikipedia corpus. Our experiments show higher correlations of the measured gender bias with the actual gender bias statistics of the U.S. job market - on two collections and with a variety of word embedding models - using the first-order approach in comparison with the vector similarity-based approaches. The first-order approach also suggests a more severe bias towards female in a few specific occupations than the other approaches.

Xiao Pu, Nikolaos Pappas, James Henderson et al.

This paper demonstrates that word sense disambiguation (WSD) can improve neural machine translation (NMT) by widening the source context considered when modeling the senses of potentially ambiguous words. We first introduce three adaptive clustering algorithms for WSD, based on k-means, Chinese restaurant processes, and random walks, which are then applied to large word contexts represented in a low-rank space and evaluated on SemEval shared-task data. We then learn word vectors jointly with sense vectors defined by our best WSD method, within a state-of-the-art NMT system. We show that the concatenation of these vectors, and the use of a sense selection mechanism based on the weighted average of sense vectors, outperforms several baselines including sense-aware ones. This is demonstrated by translation on five language pairs. The improvements are above one BLEU point over strong NMT baselines, +4% accuracy over all ambiguous nouns and verbs, or +20% when scored manually over several challenging words.

Lesly Miculicich, Dhananjay Ram, Nikolaos Pappas et al.

Neural Machine Translation (NMT) can be improved by including document-level contextual information. For this purpose, we propose a hierarchical attention model to capture the context in a structured and dynamic manner. The model is integrated in the original NMT architecture as another level of abstraction, conditioning on the NMT model's own previous hidden states. Experiments show that hierarchical attention significantly improves the BLEU score over a strong NMT baseline with the state-of-the-art in context-aware methods, and that both the encoder and decoder benefit from context in complementary ways.

Nikolaos Pappas, Lesly Miculicich Werlen, James Henderson

Tying the weights of the target word embeddings with the target word classifiers of neural machine translation models leads to faster training and often to better translation quality. Given the success of this parameter sharing, we investigate other forms of sharing in between no sharing and hard equality of parameters. In particular, we propose a structure-aware output layer which captures the semantic structure of the output space of words within a joint input-output embedding. The model is a generalized form of weight tying which shares parameters but allows learning a more flexible relationship with input word embeddings and allows the effective capacity of the output layer to be controlled. In addition, the model shares weights across output classifiers and translation contexts which allows it to better leverage prior knowledge about them. Our evaluation on English-to-Finnish and English-to-German datasets shows the effectiveness of the method against strong encoder-decoder baselines trained with or without weight tying.

Nikolaos Pappas, James Henderson

Neural text classification models typically treat output labels as categorical variables which lack description and semantics. This forces their parametrization to be dependent on the label set size, and, hence, they are unable to scale to large label sets and generalize to unseen ones. Existing joint input-label text models overcome these issues by exploiting label descriptions, but they are unable to capture complex label relationships, have rigid parametrization, and their gains on unseen labels happen often at the expense of weak performance on the labels seen during training. In this paper, we propose a new input-label model which generalizes over previous such models, addresses their limitations and does not compromise performance on seen labels. The model consists of a joint non-linear input-label embedding with controllable capacity and a joint-space-dependent classification unit which is trained with cross-entropy loss to optimize classification performance. We evaluate models on full-resource and low- or zero-resource text classification of multilingual news and biomedical text with a large label set. Our model outperforms monolingual and multilingual models which do not leverage label semantics and previous joint input-label space models in both scenarios.

James Henderson

Lexical entailment, such as hyponymy, is a fundamental issue in the semantics of natural language. This paper proposes distributional semantic models which efficiently learn word embeddings for entailment, using a recently-proposed framework for modelling entailment in a vector-space. These models postulate a latent vector for a pseudo-phrase containing two neighbouring word vectors. We investigate both modelling words as the evidence they contribute about this phrase vector, or as the posterior distribution of a one-word phrase vector, and find that the posterior vectors perform better. The resulting word embeddings outperform the best previous results on predicting hyponymy between words, in unsupervised and semi-supervised experiments.

Diana Nicoleta Popa, James Henderson

Vector-space models, from word embeddings to neural network parsers, have many advantages for NLP. But how to generalise from fixed-length word vectors to a vector space for arbitrary linguistic structures is still unclear. In this paper we propose bag-of-vector embeddings of arbitrary linguistic graphs. A bag-of-vector space is the minimal nonparametric extension of a vector space, allowing the representation to grow with the size of the graph, but not tying the representation to any specific tree or graph structure. We propose efficient training and inference algorithms based on tensor factorisation for embedding arbitrary graphs in a bag-of-vector space. We demonstrate the usefulness of this representation by training bag-of-vector embeddings of dependency graphs and evaluating them on unsupervised semantic induction for the Semantic Textual Similarity and Natural Language Inference tasks.

James Henderson, Diana Nicoleta Popa

Distributional semantics creates vector-space representations that capture many forms of semantic similarity, but their relation to semantic entailment has been less clear. We propose a vector-space model which provides a formal foundation for a distributional semantics of entailment. Using a mean-field approximation, we develop approximate inference procedures and entailment operators over vectors of probabilities of features being known (versus unknown). We use this framework to reinterpret an existing distributional-semantic model (Word2Vec) as approximating an entailment-based model of the distributions of words in contexts, thereby predicting lexical entailment relations. In both unsupervised and semi-supervised experiments on hyponymy detection, we get substantial improvements over previous results.

Nikhil Garg, James Henderson

We propose a Bayesian model of unsupervised semantic role induction in multiple languages, and use it to explore the usefulness of parallel corpora for this task. Our joint Bayesian model consists of individual models for each language plus additional latent variables that capture alignments between roles across languages. Because it is a generative Bayesian model, we can do evaluations in a variety of scenarios just by varying the inference procedure, without changing the model, thereby comparing the scenarios directly. We compare using only monolingual data, using a parallel corpus, using a parallel corpus with annotations in the other language, and using small amounts of annotation in the target language. We find that the biggest impact of adding a parallel corpus to training is actually the increase in mono-lingual data, with the alignments to another language resulting in small improvements, even with labeled data for the other language.

Joel Lang, James Henderson

Previous work has shown the effectiveness of random walk hitting times as a measure of dissimilarity in a variety of graph-based learning problems such as collaborative filtering, query suggestion or finding paraphrases. However, application of hitting times has been limited to small datasets because of computational restrictions. This paper develops a new approximation algorithm with which hitting times can be computed on very large, disk-resident graphs, making their application possible to problems which were previously out of reach. This will potentially benefit a range of large-scale problems.