A possible explanation for the impressive performance of masked language model (MLM) pre-training is that such models have learned to represent the syntactic structures prevalent in classical NLP pipelines. In this paper, we propose a different expla
nation: MLMs succeed on downstream tasks almost entirely due to their ability to model higher-order word co-occurrence statistics. To demonstrate this, we pre-train MLMs on sentences with randomly shuffled word order, and show that these models still achieve high accuracy after fine-tuning on many downstream tasks---including tasks specifically designed to be challenging for models that ignore word order. Our models perform surprisingly well according to some parametric syntactic probes, indicating possible deficiencies in how we test representations for syntactic information. Overall, our results show that purely distributional information largely explains the success of pre-training, and underscore the importance of curating challenging evaluation datasets that require deeper linguistic knowledge.
Reading order detection is the cornerstone to understanding visually-rich documents (e.g., receipts and forms). Unfortunately, no existing work took advantage of advanced deep learning models because it is too laborious to annotate a large enough dat
aset. We observe that the reading order of WORD documents is embedded in their XML metadata; meanwhile, it is easy to convert WORD documents to PDFs or images. Therefore, in an automated manner, we construct ReadingBank, a benchmark dataset that contains reading order, text, and layout information for 500,000 document images covering a wide spectrum of document types. This first-ever large-scale dataset unleashes the power of deep neural networks for reading order detection. Specifically, our proposed LayoutReader captures the text and layout information for reading order prediction using the seq2seq model. It performs almost perfectly in reading order detection and significantly improves both open-source and commercial OCR engines in ordering text lines in their results in our experiments. The dataset and models are publicly available at https://aka.ms/layoutreader.
Undirected neural sequence models have achieved performance competitive with the state-of-the-art directed sequence models that generate monotonically from left to right in machine translation tasks. In this work, we train a policy that learns the ge
neration order for a pre-trained, undirected translation model via reinforcement learning. We show that the translations decoded by our learned orders achieve higher BLEU scores than the outputs decoded from left to right or decoded by the learned order from Mansimov et al. (2019) on the WMT'14 German-English translation task. On examples with a maximum source and target length of 30 from De-En and WMT'16 English-Romanian tasks, our learned order outperforms all heuristic generation orders on three out of four language pairs. We next carefully analyze the learned order patterns via qualitative and quantitative analysis. We show that our policy generally follows an outer-to-inner order, predicting the left-most and right-most positions first, and then moving toward the middle while skipping less important words at the beginning. Furthermore, the policy usually predicts positions for a single syntactic constituent structure in consecutive steps. We believe our findings could provide more insights on the mechanism of undirected generation models and encourage further research in this direction.
Tag recommendation relies on either a ranking function for top-k tags or an autoregressive generation method. However, the previous methods neglect one of two seemingly conflicting yet desirable characteristics of a tag set: orderlessness and inter-d
ependency. While the ranking approach fails to address the inter-dependency among tags when they are ranked, the autoregressive approach fails to take orderlessness into account because it is designed to utilize sequential relations among tokens. We propose a sequence-oblivious generation method for tag recommendation, in which the next tag to be generated is independent of the order of the generated tags and the order of the ground truth tags occurring in training data. Empirical results on two different domains, Instagram and Stack Overflow, show that our method is significantly superior to the previous approaches.
Recently, language models (LMs) have achieved significant performance on many NLU tasks, which has spurred widespread interest for their possible applications in the scientific and social area. However, LMs have faced much criticism of whether they a
re truly capable of reasoning in NLU. In this work, we propose a diagnostic method for first-order logic (FOL) reasoning with a new proposed benchmark, LogicNLI. LogicNLI is an NLI-style dataset that effectively disentangles the target FOL reasoning from commonsense inference and can be used to diagnose LMs from four perspectives: accuracy, robustness, generalization, and interpretability. Experiments on BERT, RoBERTa, and XLNet, have uncovered the weaknesses of these LMs on FOL reasoning, which motivates future exploration to enhance the reasoning ability.
Most natural languages have a predominant or fixed word order. For example in English the word order is usually Subject-Verb-Object. This work attempts to explain this phenomenon as well as other typological findings regarding word order from a funct
ional perspective. In particular, we examine whether fixed word order provides a functional advantage, explaining why these languages are prevalent. To this end, we consider an evolutionary model of language and demonstrate, both theoretically and using genetic algorithms, that a language with a fixed word order is optimal. We also show that adding information to the sentence, such as case markers and noun-verb distinction, reduces the need for fixed word order, in accordance with the typological findings.
This paper details experiments we performed on the Universal Dependencies 2.7 corpora in order to investigate the dominant word order in the available languages. For this purpose, we used a graph rewriting tool, GREW, which allowed us to go beyond th
e surface annotations and identify the implicit subjects. We first measured the distribution of the six different word orders (SVO, SOV, VSO, VOS, OVS, OSV) in the corpora and investigated when there was a significant difference in the corpora within a given language. Then, we compared the obtained results with information provided in the WALS database (Dryer and Haspelmath, 2013) and in ( ̈Ostling, 2015). Finally, we examined the impact of using a graph rewriting tool for this task. The tools and resources used for this research are all freely available.
We observe an instance of gender-induced bias in a downstream application, despite the absence of explicit gender words in the test cases. We provide a test set, SoWinoBias, for the purpose of measuring such latent gender bias in coreference resoluti
on systems. We evaluate the performance of current debiasing methods on the SoWinoBias test set, especially in reference to the method's design and altered embedding space properties. See https://github.com/hillary-dawkins/SoWinoBias.
Domain divergence plays a significant role in estimating the performance of a model in new domains. While there is a significant literature on divergence measures, researchers find it hard to choose an appropriate divergence for a given NLP applicati
on. We address this shortcoming by both surveying the literature and through an empirical study. We develop a taxonomy of divergence measures consisting of three classes --- Information-theoretic, Geometric, and Higher-order measures and identify the relationships between them. Further, to understand the common use-cases of these measures, we recognise three novel applications -- 1) Data Selection, 2) Learning Representation, and 3) Decisions in the Wild -- and use it to organise our literature. From this, we identify that Information-theoretic measures are prevalent for 1) and 3), and Higher-order measures are more common for 2). To further help researchers choose appropriate measures to predict drop in performance -- an important aspect of Decisions in the Wild, we perform correlation analysis spanning 130 domain adaptation scenarios, 3 varied NLP tasks and 12 divergence measures identified from our survey. To calculate these divergences, we consider the current contextual word representations (CWR) and contrast with the older distributed representations. We find that traditional measures over word distributions still serve as strong baselines, while higher-order measures with CWR are effective.
First-order meta-learning algorithms have been widely used in practice to learn initial model parameters that can be quickly adapted to new tasks due to their efficiency and effectiveness. However, existing studies find that meta-learner can overfit
to some specific adaptation when we have heterogeneous tasks, leading to significantly degraded performance. In Natural Language Processing (NLP) applications, datasets are often diverse and each task has its unique characteristics. Therefore, to address the overfitting issue when applying first-order meta-learning to NLP applications, we propose to reduce the variance of the gradient estimator used in task adaptation. To this end, we develop a variance-reduced first-order meta-learning algorithm. The core of our algorithm is to introduce a novel variance reduction term to the gradient estimation when performing the task adaptation. Experiments on two NLP applications: few-shot text classification and multi-domain dialog state tracking demonstrate the superior performance of our proposed method.
