Transformer architectures have achieved state- of-the-art results on a variety of natural language processing (NLP) tasks. However, their attention mechanism comes with a quadratic complexity in sequence lengths, making the computational overhead prohibitive, especially for long sequences. Attention context can be seen as a random-access memory with each token taking a slot. Under this perspective, the memory size grows linearly with the sequence length, and so does the overhead of reading from it. One way to improve the efficiency is to bound the memory size. We show that disparate approaches can be subsumed into one abstraction, attention with bounded-memory control (ABC), and they vary in their organization of the memory. ABC reveals new, unexplored possibilities. First, it connects several efficient attention variants that would otherwise seem apart. Second, this abstraction gives new insights—an established approach (Wang et al., 2020b) previously thought to not be applicable in causal attention, actually is. Last, we present a new instance of ABC, which draws inspiration from existing ABC approaches, but replaces their heuristic memory-organizing functions with a learned, contextualized one. Our experiments on language modeling, machine translation, and masked language model finetuning show that our approach outperforms previous efficient attention models; compared to the strong transformer baselines, it significantly improves the inference time and space efficiency with no or negligible accuracy loss.

We propose knowledge internalization (KI), which aims to complement the lexical knowledge into neural dialog models. Instead of further conditioning the knowledge-grounded dialog (KGD) models on externally retrieved knowledge, we seek to integrate knowledge about each input token internally into the model’s parameters. To tackle the challenge due to the large scale of lexical knowledge, we adopt the contrastive learning approach and create an effective token-level lexical knowledge retriever that requires only weak supervision mined from Wikipedia. We demonstrate the effectiveness and general applicability of our approach on various datasets and diversified model structures.

Open-ended text generation tasks, such as dialogue generation and story completion, require models to generate a coherent continuation given limited preceding context. The open-ended nature of these tasks brings new challenges to the neural auto-regressive text generators nowadays. Despite these neural models are good at producing human-like text, it is difficult for them to arrange causalities and relations between given facts and possible ensuing events. To bridge this gap, we propose a novel two-stage method which explicitly arranges the ensuing events in open-ended text generation. Our approach can be understood as a specially-trained coarse-to-fine algorithm, where an event transition planner provides a “coarse” plot skeleton and a text generator in the second stage refines the skeleton. Experiments on two open-ended text generation tasks demonstrate that our proposed method effectively improves the quality of the generated text, especially in coherence and diversity. We will release the codes to the community for further exploration.

We examine the extent to which, in principle, different syntactic and semantic graph representations can complement and improve neural language modeling. Specifically, by conditioning on a subgraph encapsulating the locally relevant sentence history, can a model make better next-word predictions than a pretrained sequential language model alone? With an ensemble setup consisting of GPT-2 and ground-truth graphs from one of 7 different formalisms, we find that the graph information indeed improves perplexity and other metrics. Moreover, this architecture provides a new way to compare different frameworks of linguistic representation. In our oracle graph setup, training and evaluating on English WSJ, semantic constituency structures prove most useful to language modeling performance—outpacing syntactic constituency structures as well as syntactic and semantic dependency structures.

Transformers have advanced the field of natural language processing (NLP) on a variety of important tasks. At the cornerstone of the Transformer architecture is the multi-head attention (MHA) mechanism which models pairwise interactions between the elements of the sequence. Despite its massive success, the current framework ignores interactions among different heads, leading to the problem that many of the heads are redundant in practice, which greatly wastes the capacity of the model. To improve parameter efficiency, we re-formulate the MHA as a latent variable model from a probabilistic perspective. We present cascaded head-colliding attention (CODA) which explicitly models the interactions between attention heads through a hierarchical variational distribution. We conduct extensive experiments and demonstrate that CODA outperforms the transformer baseline, by 0.6 perplexity on Wikitext-103 in language modeling, and by 0.6 BLEU on WMT14 EN-DE in machine translation, due to its improvements on the parameter efficiency.

A neural multimodal machine translation (MMT) system is one that aims to perform better translation by extending conventional text-only translation models with multimodal information. Many recent studies report improvements when equipping their models with the multimodal module, despite the controversy of whether such improvements indeed come from the multimodal part. We revisit the contribution of multimodal information in MMT by devising two interpretable MMT models. To our surprise, although our models replicate similar gains as recently developed multimodal-integrated systems achieved, our models learn to ignore the multimodal information. Upon further investigation, we discover that the improvements achieved by the multimodal models over text-only counterparts are in fact results of the regularization effect. We report empirical findings that highlight the importance of MMT models’ interpretability, and discuss how our findings will benefit future research.

Abstract We present a language model that combines a large parametric neural network (i.e., a transformer) with a non-parametric episodic memory component in an integrated architecture. Our model uses extended short-term context by caching local hidden states—similar to transformer-XL—and global long-term memory by retrieving a set of nearest neighbor tokens at each timestep. We design a gating function to adaptively combine multiple information sources to make a prediction. This mechanism allows the model to use either local context, short-term memory, or long-term memory (or any combination of them) on an ad hoc basis depending on the context. Experiments on word-based and character-based language modeling datasets demonstrate the efficacy of our proposed method compared to strong baselines.

We show that Bayes’ rule provides an effective mechanism for creating document translation models that can be learned from only parallel sentences and monolingual documents a compelling benefit because parallel documents are not always available. In our formulation, the posterior probability of a candidate translation is the product of the unconditional (prior) probability of the candidate output document and the “reverse translation probability” of translating the candidate output back into the source language. Our proposed model uses a powerful autoregressive language model as the prior on target language documents, but it assumes that each sentence is translated independently from the target to the source language. Crucially, at test time, when a source document is observed, the document language model prior induces dependencies between the translations of the source sentences in the posterior. The model’s independence assumption not only enables efficient use of available data, but it additionally admits a practical left-to-right beam-search algorithm for carrying out inference. Experiments show that our model benefits from using cross-sentence context in the language model, and it outperforms existing document translation approaches.

Textual representation learners trained on large amounts of data have achieved notable success on downstream tasks; intriguingly, they have also performed well on challenging tests of syntactic competence. Hence, it remains an open question whether scalable learners like BERT can become fully proficient in the syntax of natural language by virtue of data scale alone, or whether they still benefit from more explicit syntactic biases. To answer this question, we introduce a knowledge distillation strategy for injecting syntactic biases into BERT pretraining, by distilling the syntactically informative predictions of a hierarchical—albeit harder to scale—syntactic language model. Since BERT models masked words in bidirectional context, we propose to distill the approximate marginal distribution over words in context from the syntactic LM. Our approach reduces relative error by 2–21% on a diverse set of structured prediction tasks, although we obtain mixed results on the GLUE benchmark. Our findings demonstrate the benefits of syntactic biases, even for representation learners that exploit large amounts of data, and contribute to a better understanding of where syntactic biases are helpful in benchmarks of natural language understanding.

Recurrent neural network grammars (RNNG) are a recently proposed probablistic generative modeling family for natural language. They show state-of-the-art language modeling and parsing performance. We investigate what information they learn, from a linguistic perspective, through various ablations to the model and the data, and by augmenting the model with an attention mechanism (GA-RNNG) to enable closer inspection. We find that explicit modeling of composition is crucial for achieving the best performance. Through the attention mechanism, we find that headedness plays a central role in phrasal representation (with the model’s latent attention largely agreeing with predictions made by hand-crafted head rules, albeit with some important differences). By training grammars without nonterminal labels, we find that phrasal representations depend minimally on nonterminals, providing support for the endocentricity hypothesis.