Social biases such as gender or racial biases have been reported in language models (LMs), including Masked Language Models (MLMs). Given that MLMs are continuously trained with increasing amounts of additional data collected over time, an important yet unanswered question is how the social biases encoded with MLMs vary over time. In particular, the number of social media users continues to grow at an exponential rate, and it is a valid concern for the MLMs trained specifically on social media data whether their social biases (if any) would also amplify over time. To empirically analyse this problem, we use a series of MLMs pretrained on chronologically ordered temporal snapshots of corpora. Our analysis reveals that, although social biases are present in all MLMs, most types of social bias remain relatively stable over time (with a few exceptions). To further understand the mechanisms that influence social biases in MLMs, we analyse the temporal corpora used to train the MLMs. Our findings show that some demographic groups, such as male, obtain higher preference over the other, such as female on the training corpora constantly.

Sentence embeddings produced by Pretrained Language Models (PLMs) have received wide attention from the NLP community due to their superior performance when representing texts in numerous downstream applications. However, the high dimensionality of the sentence embeddings produced by PLMs is problematic when representing large numbers of sentences in memory- or compute-constrained devices. As a solution, we evaluate unsupervised dimensionality reduction methods to reduce the dimensionality of sentence embeddings produced by PLMs. Our experimental results show that simple methods such as Principal Component Analysis (PCA) can reduce the dimensionality of sentence embeddings by almost 50%, without incurring a significant loss in performance in multiple downstream tasks. Surprisingly, reducing the dimensionality further improves performance over the original high dimensional versions for the sentence embeddings produced by some PLMs in some tasks.

Dynamic contextualised word embeddings (DCWEs) represent the temporal semantic variations of words. We propose a method for learning DCWEs by time-adapting a pretrained Masked Language Model (MLM) using time-sensitive templates. Given two snapshots C₁ and C₂ of a corpus taken respectively at two distinct timestamps T₁ and T₂, we first propose an unsupervised method to select (a) pivot terms related to both C₁ and C₂, and (b) anchor terms that are associated with a specific pivot term in each individual snapshot.We then generate prompts by filling manually compiled templates using the extracted pivot and anchor terms.Moreover, we propose an automatic method to learn time-sensitive templates from C₁ and C₂, without requiring any human supervision.Next, we use the generated prompts to adapt a pretrained MLM to T₂ by fine-tuning using those prompts.Multiple experiments show that our proposed method significantly reduces the perplexity of test sentences in C₂, outperforming the current state-of-the-art.

Natural language is ambiguous. Resolving ambiguous questions is key to successfully answering them. Focusing on questions about images, we create a dataset of ambiguous examples. We annotate these, grouping answers by the underlying question they address and rephrasing the question for each group to reduce ambiguity. Our analysis reveals a linguistically-aligned ontology of reasons for ambiguity in visual questions. We then develop an English question-generation model which we demonstrate via automatic and human evaluation produces less ambiguous questions. We further show that the question generation objective we use allows the model to integrate answer group information without any direct supervision.

Sense embedding learning methods learn multiple vectors for a given ambiguous word, corresponding to its different word senses. For this purpose, different methods have been proposed in prior work on sense embedding learning that use different sense inventories, sense-tagged corpora and learning methods. However, not all existing sense embeddings cover all senses of ambiguous words equally well due to the discrepancies in their training resources. To address this problem, we propose the first-ever meta-sense embedding method – Neighbour Preserving Meta-Sense Embeddings, which learns meta-sense embeddings by combining multiple independently trained source sense embeddings such that the sense neighbourhoods computed from the source embeddings are preserved in the meta-embedding space. Our proposed method can combine source sense embeddings that cover different sets of word senses. Experimental results on Word Sense Disambiguation (WSD) and Word-in-Context (WiC) tasks show that the proposed meta-sense embedding method consistently outperforms several competitive baselines. An anonymised version of the source code implementation for our proposed method is submitted to reviewing system. Both source code and the learnt meta-sense embeddings will be publicly released upon paper acceptance.

Cosine similarity between two words, computed using their contextualised token embeddings obtained from masked language models (MLMs) such as BERT has shown to underestimate the actual similarity between those words CITATION.This similarity underestimation problem is particularly severe for high frequent words. Although this problem has been noted in prior work, no solution has been proposed thus far. We observe that the ℓ₂ norm of contextualised embeddings of a word correlates with its log-frequency in the pretraining corpus.Consequently, the larger ℓ₂ norms associated with the high frequent words reduce the cosine similarity values measured between them, thus underestimating the similarity scores.To solve this issue, we propose a method to discount the ℓ₂ norm of a contextualised word embedding by the frequency of that word in a corpus when measuring the cosine similarities between words.We show that the so called stop words behave differently from the rest of the words, which require special consideration during their discounting process.Experimental results on a contextualised word similarity dataset show that our proposed discounting method accurately solves the similarity underestimation problem.An anonymized version of the source code of our proposed method is submitted to the reviewing system.

In this work, we argue that non-autoregressive (NAR) sequence generative models can equivalently be regarded as an iterative refinement process towards the target sequence, implying an underlying dynamical system of NAR model: z = f (z, x) → y. In such a way, the optimal prediction of a NAR model should be the equilibrium state of its dynamics if given infinitely many iterations. However, this is infeasible in practice due to limited computational and memory budgets. To this end, we propose DEQNAR to directly solve for the equilibrium state of NAR models based on deep equilibrium networks (Bai et al., 2019) with black-box root-finding solvers and back-propagate through the equilibrium point via implicit differentiation with constant memory. We conduct extensive experiments on four WMT machine translation benchmarks. Our main findings show that DEQNAR can indeed converge to a more accurate prediction and is a general-purpose framework that consistently helps yield substantial improvement for several strong NAR backbones.

Semantic Change Detection of words is an important task for various NLP applications that must make time-sensitive predictions. Some words are used over time in novel ways to express new meanings, and these new meanings establish themselves as novel senses of existing words. On the other hand, Word Sense Disambiguation (WSD) methods associate ambiguous words with sense ids, depending on the context in which they occur. Given this relationship between WSD and SCD, we explore the possibility of predicting whether a target word has its meaning changed between two corpora collected at different time steps, by comparing the distributions of senses of that word in each corpora. For this purpose, we use pretrained static sense embeddings to automatically annotate each occurrence of the target word in a corpus with a sense id. Next, we compute the distribution of sense ids of a target word in a given corpus. Finally, we use different divergence or distance measures to quantify the semantic change of the target word across the two given corpora. Our experimental results on SemEval 2020 Task 1 dataset show that word sense distributions can be accurately used to predict semantic changes of words in English, German, Swedish and Latin.

Large-scale pre-training is widely used in recent document understanding tasks. During deployment, one may expect that models should trigger a conservative fallback policy when encountering out-of-distribution (OOD) samples, which highlights the importance of OOD detection. However, most existing OOD detection methods focus on single-modal inputs such as images or texts. While documents are multi-modal in nature, it is underexplored if and how multi-modal information in documents can be exploited for OOD detection. In this work, we first provide a systematic and in-depth analysis on OOD detection for document understanding models. We study the effects of model modality, pre-training, and fine-tuning across various types of OOD inputs. In particular, we find that spatial information is critical for document OOD detection. To better exploit spatial information, we propose a spatial-aware adapter, which serves as a parameter-efficient add-on module to adapt transformer-based language models to the document domain. Extensive experiments show that adding the spatial-aware adapter significantly improves the OOD detection performance compared to directly using the language model and achieves superior performance compared to competitive baselines.

Mathematical understanding and reasoning are crucial tasks for assessing the capabilities of artificial intelligence (AI). However, existing benchmarks either require just a few steps of reasoning, or only contain a small amount of data in one specific topic, making it hard to analyse AI’s behaviour with reference to different problems within a specific topic in detail. In this work, we propose Conic10K, a challenging math problem dataset on conic sections in Chinese senior high school education. Our dataset contains various problems with different reasoning depths, while only the knowledge from conic sections is required. Since the dataset only involves a narrow range of knowledge, it is easy to separately analyse the knowledge a model possesses and the reasoning ability it has. For each problem, we provide a high-quality formal representation, the reasoning steps, and the final solution. Experiments show that existing large language models, including GPT-4, exhibit weak performance on complex reasoning. We hope that our findings could inspire more advanced techniques for precise natural language understanding and reasoning. Our dataset and codes are available at https://github.com/whyNLP/Conic10K.

Multilingual language models (LMs) have become a powerful tool in NLP, especially for non-English languages. Nevertheless, model parameters of multilingual LMs remain large due to the larger embedding matrix of the vocabulary covering tokens in different languages. Instead, monolingual LMs can be trained in a target language with the language-specific vocabulary only. In this paper, we propose vocabulary-trimming (VT), a method to reduce a multilingual LM vocabulary to a target language by deleting potentially irrelevant tokens from its vocabulary. In theory, VT can compress any existing multilingual LM to any language covered by the original model. In our experiments, we show that VT can retain the original performance of the multilingual LM, while being considerably smaller in size than the original multilingual LM. The evaluation is performed over four NLP tasks (two generative and two classification tasks) among four widely used multilingual LMs in seven languages. The results show that this methodology can keep the best of both monolingual and multilingual worlds by keeping a small size as monolingual models without the need for specifically retraining them, and can even help limit potentially harmful social biases.

Handwritten mathematical expression recognition (HMER) is a multidisciplinary task that generates LaTeX sequences from images. Existing approaches, employing tree decoders within attention-based encoder-decoder architectures, aim to capture the hierarchical tree structure, but are limited by CFGs and pre-generated triplet data, hindering expandability and neglecting visual ambiguity challenges. This article investigates the distinctive language characteristics of LaTeX mathematical expressions, revealing two key observations: 1) the presence of explicit structural symbols, and 2) the treatment of symbols, particularly letters, as minimal units with context-dependent semantics, representing variables or constants. Rooted in these properties, we propose that language models have the potential to synchronously and complementarily provide both structural and semantic information, making them suitable for correction of HMER. To validate our proposition, we propose an architecture called Recognize and Language Fusion Network (RLFN), which integrates recognition and language features to output corrected sequences while jointly optimizing with a string decoder recognition model. Experiments show that RLFN outperforms existing state-of-the-art methods on the CROHME 2014/2016/2019 datasets.

Various types of social biases have been reported with pretrained Masked Language Models (MLMs) in prior work. However, multiple underlying factors are associated with an MLM such as its model size, size of the training data, training objectives, the domain from which pretraining data is sampled, tokenization, and languages present in the pretrained corpora, to name a few. It remains unclear as to which of those factors influence social biases that are learned by MLMs. To study the relationship between model factors and the social biases learned by an MLM, as well as the downstream task performance of the model, we conduct a comprehensive study over 39 pretrained MLMs covering different model sizes, training objectives, tokenization methods, training data domains and languages. Our results shed light on important factors often neglected in prior literature, such as tokenization or model objectives.

Sense embedding learning methods learn different embeddings for the different senses of an ambiguous word. One sense of an ambiguous word might be socially biased while its other senses remain unbiased. In comparison to the numerous prior work evaluating the social biases in pretrained word embeddings, the biases in sense embeddings have been relatively understudied. We create a benchmark dataset for evaluating the social biases in sense embeddings and propose novel sense-specific bias evaluation measures. We conduct an extensive evaluation of multiple static and contextualised sense embeddings for various types of social biases using the proposed measures. Our experimental results show that even in cases where no biases are found at word-level, there still exist worrying levels of social biases at sense-level, which are often ignored by the word-level bias evaluation measures.

We show that the l2 norm of a static sense embedding encodes information related to the frequency of that sense in the training corpus used to learn the sense embeddings. This finding can be seen as an extension of a previously known relationship for word embeddings to sense embeddings. Our experimental results show that in spite of its simplicity, the l2 norm of sense embeddings is a surprisingly effective feature for several word sense related tasks such as (a) most frequent sense prediction, (b) word-in-context (WiC), and (c) word sense disambiguation (WSD). In particular, by simply including the l2 norm of a sense embedding as a feature in a classifier, we show that we can improve WiC and WSD methods that use static sense embeddings.

Automatic word segmentation and part-of-speech tagging of ancient books can help relevant researchers to study ancient texts. In recent years, pre-trained language models have achieved significant improvements on text processing tasks. SikuRoberta is a pre-trained language model specially designed for automatic analysis of ancient Chinese texts. Although SikuRoberta significantly boosts performance on WSG and POS tasks on ancient Chinese texts, the lack of labeled data still limits the performance of the model. In this paper, to alleviate the problem of insufficient training data, We define hybrid tags to integrate WSG and POS tasks and design Roberta-CRF model to predict tags for each Chinese characters. Moreover, We generate synthetic labeled data based on the LSTM language model. To further mine knowledge in SikuRoberta, we generate the synthetic unlabeled data based on the Masked LM. Experiments show that the performance of the model is improved with the synthetic data, indicating that the effectiveness of the data augmentation methods.

Although deep neural networks have achieved prominent performance on many NLP tasks, they are vulnerable to adversarial examples. We propose Dirichlet Neighborhood Ensemble (DNE), a randomized method for training a robust model to defense synonym substitution-based attacks. During training, DNE forms virtual sentences by sampling embedding vectors for each word in an input sentence from a convex hull spanned by the word and its synonyms, and it augments them with the training data. In such a way, the model is robust to adversarial attacks while maintaining the performance on the original clean data. DNE is agnostic to the network architectures and scales to large models (e.g., BERT) for NLP applications. Through extensive experimentation, we demonstrate that our method consistently outperforms recently proposed defense methods by a significant margin across different network architectures and multiple data sets.

This paper describes the Volctrans’ submission to the WMT21 news translation shared task for German->English translation. We build a parallel (i.e., non-autoregressive) translation system using the Glancing Transformer, which enables fast and accurate parallel decoding in contrast to the currently prevailing autoregressive models. To the best of our knowledge, this is the first parallel translation system that can be scaled to such a practical scenario like WMT competition. More importantly, our parallel translation system achieves the best BLEU score (35.0) on German->English translation task, outperforming all strong autoregressive counterparts.

Deep neural networks are vulnerable to adversarial attacks, where a small perturbation to an input alters the model prediction. In many cases, malicious inputs intentionally crafted for one model can fool another model. In this paper, we present the first study to systematically investigate the transferability of adversarial examples for text classification models and explore how various factors, including network architecture, tokenization scheme, word embedding, and model capacity, affect the transferability of adversarial examples. Based on these studies, we propose a genetic algorithm to find an ensemble of models that can be used to induce adversarial examples to fool almost all existing models. Such adversarial examples reflect the defects of the learning process and the data bias in the training set. Finally, we derive word replacement rules that can be used for model diagnostics from these adversarial examples.

Despite achieving prominent performance on many important tasks, it has been reported that neural networks are vulnerable to adversarial examples. Previously studies along this line mainly focused on semantic tasks such as sentiment analysis, question answering and reading comprehension. In this study, we show that adversarial examples also exist in dependency parsing: we propose two approaches to study where and how parsers make mistakes by searching over perturbations to existing texts at sentence and phrase levels, and design algorithms to construct such examples in both of the black-box and white-box settings. Our experiments with one of state-of-the-art parsers on the English Penn Treebank (PTB) show that up to 77% of input examples admit adversarial perturbations, and we also show that the robustness of parsing models can be improved by crafting high-quality adversaries and including them in the training stage, while suffering little to no performance drop on the clean input data.

Entity embeddings, which represent different aspects of each entity with a single vector like word embeddings, are a key component of neural entity linking models. Existing entity embeddings are learned from canonical Wikipedia articles and local contexts surrounding target entities. Such entity embeddings are effective, but too distinctive for linking models to learn contextual commonality. We propose a simple yet effective method, FGS2EE, to inject fine-grained semantic information into entity embeddings to reduce the distinctiveness and facilitate the learning of contextual commonality. FGS2EE first uses the embeddings of semantic type words to generate semantic embeddings, and then combines them with existing entity embeddings through linear aggregation. Extensive experiments show the effectiveness of such embeddings. Based on our entity embeddings, we achieved new sate-of-the-art performance on entity linking.

We propose a novel approach to cross-lingual dependency parsing based on word reordering. The words in each sentence of a source language corpus are rearranged to meet the word order in a target language under the guidance of a part-of-speech based language model (LM). To obtain the highest reordering score under the LM, a population-based optimization algorithm and its genetic operators are designed to deal with the combinatorial nature of such word reordering. A parser trained on the reordered corpus then can be used to parse sentences in the target language. We demonstrate through extensive experimentation that our approach achieves better or comparable results across 25 target languages (1.73% increase in average), and outperforms a baseline by a significant margin on the languages that are greatly different from the source one. For example, when transferring the English parser to Hindi and Latin, our approach outperforms the baseline by 15.3% and 6.7% respectively.

AiFu has won the first place in the SemEval-2019 Task 10 - ”Math Question Answering”competition. This paper is to describe how it works technically and to report and analyze some essential experimental results

The Biological Text Mining Unit at BSC and CNIO organized the first shared task on chemical & drug mention recognition from Spanish medical texts called PharmaCoNER (Pharmacological Substances, Compounds and proteins and Named Entity Recognition track) in 2019, which includes two tracks: one for NER offset and entity classification (track 1) and the other one for concept indexing (track 2). We developed a pipeline system based on deep learning methods for this shared task, specifically, a subsystem based on BERT (Bidirectional Encoder Representations from Transformers) for NER offset and entity classification and a subsystem based on Bpool (Bi-LSTM with max/mean pooling) for concept indexing. Evaluation conducted on the shared task data showed that our system achieves a micro-average F1-score of 0.9105 on track 1 and a micro-average F1-score of 0.8391 on track 2.

Discourse parsing is a challenging task and plays a critical role in discourse analysis. This paper focus on the macro level discourse structure analysis, which has been less studied in the previous researches. We explore a macro discourse structure presentation schema to present the macro level discourse structure, and propose a corresponding corpus, named Macro Chinese Discourse Treebank. On these bases, we concentrate on two tasks of macro discourse structure analysis, including structure identification and nuclearity recognition. In order to reduce the error transmission between the associated tasks, we adopt a joint model of the two tasks, and an Integer Linear Programming approach is proposed to achieve global optimization with various kinds of constraints.