Dense retrieval methods have shown great promise over sparse retrieval methods in a range of NLP problems. Among them, dense phrase retrieval-the most fine-grained retrieval unit-is appealing because phrases can be directly used as the output for que
stion answering and slot filling tasks. In this work, we follow the intuition that retrieving phrases naturally entails retrieving larger text blocks and study whether phrase retrieval can serve as the basis for coarse-level retrieval including passages and documents. We first observe that a dense phrase-retrieval system, without any retraining, already achieves better passage retrieval accuracy (+3-5% in top-5 accuracy) compared to passage retrievers, which also helps achieve superior end-to-end QA performance with fewer passages. Then, we provide an interpretation for why phrase-level supervision helps learn better fine-grained entailment compared to passage-level supervision, and also show that phrase retrieval can be improved to achieve competitive performance in document-retrieval tasks such as entity linking and knowledge-grounded dialogue. Finally, we demonstrate how phrase filtering and vector quantization can reduce the size of our index by 4-10x, making dense phrase retrieval a practical and versatile solution in multi-granularity retrieval.
The creation of a large summarization quality dataset is a considerable, expensive, time-consuming effort, requiring careful planning and setup. It includes producing human-written and machine-generated summaries and evaluation of the summaries by hu
mans, preferably by linguistic experts, and by automatic evaluation tools. If such effort is made in one language, it would be beneficial to be able to use it in other languages. To investigate how much we can trust the translation of such dataset without repeating human annotations in another language, we translated an existing English summarization dataset, SummEval dataset, to four different languages and analyzed the scores from the automatic evaluation metrics in translated languages, as well as their correlation with human annotations in the source language. Our results reveal that although translation changes the absolute value of automatic scores, the scores keep the same rank order and approximately the same correlations with human annotations.
We compare two orthogonal semi-supervised learning techniques, namely tri-training and pretrained word embeddings, in the task of dependency parsing. We explore language-specific FastText and ELMo embeddings and multilingual BERT embeddings. We focus
on a low resource scenario as semi-supervised learning can be expected to have the most impact here. Based on treebank size and available ELMo models, we select Hungarian, Uyghur (a zero-shot language for mBERT) and Vietnamese. Furthermore, we include English in a simulated low-resource setting. We find that pretrained word embeddings make more effective use of unlabelled data than tri-training but that the two approaches can be successfully combined.
The societal issue of digital hostility has previously attracted a lot of attention. The topic counts an ample body of literature, yet remains prominent and challenging as ever due to its subjective nature. We posit that a better understanding of thi
s problem will require the use of causal inference frameworks. This survey summarises the relevant research that revolves around estimations of causal effects related to online hate speech. Initially, we provide an argumentation as to why re-establishing the exploration of hate speech in causal terms is of the essence. Following that, we give an overview of the leading studies classified with respect to the direction of their outcomes, as well as an outline of all related research, and a summary of open research problems that can influence future work on the topic.
Much of natural language processing is focused on leveraging large capacity language models, typically trained over single messages with a task of predicting one or more tokens. However, modeling human language at higher-levels of context (i.e., sequ
ences of messages) is under-explored. In stance detection and other social media tasks where the goal is to predict an attribute of a message, we have contextual data that is loosely semantically connected by authorship. Here, we introduce Message-Level Transformer (MeLT) -- a hierarchical message-encoder pre-trained over Twitter and applied to the task of stance prediction. We focus on stance prediction as a task benefiting from knowing the context of the message (i.e., the sequence of previous messages). The model is trained using a variant of masked-language modeling; where instead of predicting tokens, it seeks to generate an entire masked (aggregated) message vector via reconstruction loss. We find that applying this pre-trained masked message-level transformer to the downstream task of stance detection achieves F1 performance of 67%.
Typically, Open Information Extraction (OpenIE) focuses on extracting triples, representing a subject, a relation, and the object of the relation. However, most of the existing techniques are based on a predefined set of relations in each domain whic
h limits their applicability to newer domains where these relations may be unknown such as financial documents. This paper presents a zero-shot open information extraction technique that extracts the entities (value) and their descriptions (key) from a sentence, using off the shelf machine reading comprehension (MRC) Model. The input questions to this model are created using a novel noun phrase generation method. This method takes the context of the sentence into account and can create a wide variety of questions making our technique domain independent. Given the questions and the sentence, our technique uses the MRC model to extract entities (value). The noun phrase corresponding to the question, with the highest confidence, is taken as the description (key). This paper also introduces the EDGAR10-Q dataset which is based on publicly available financial documents from corporations listed in US securities and exchange commission (SEC). The dataset consists of paragraphs, tagged values (entities), and their keys (descriptions) and is one of the largest among entity extraction datasets. This dataset will be a valuable addition to the research community, especially in the financial domain. Finally, the paper demonstrates the efficacy of the proposed technique on the EDGAR10-Q and Ade corpus drug dosage datasets, where it obtained 86.84 % and 97% accuracy, respectively.
We present ARETA, an automatic error type annotation system for Modern Standard Arabic. We design ARETA to address Arabics morphological richness and orthographic ambiguity. We base our error taxonomy on the Arabic Learner Corpus (ALC) Error Tagset w
ith some modifications. ARETA achieves a performance of 85.8% (micro average F1 score) on a manually annotated blind test portion of ALC. We also demonstrate ARETAs usability by applying it to a number of submissions from the QALB 2014 shared task for Arabic grammatical error correction. The resulting analyses give helpful insights on the strengths and weaknesses of different submissions, which is more useful than the opaque M2 scoring metrics used in the shared task. ARETA employs a large Arabic morphological analyzer, but is completely unsupervised otherwise. We make ARETA publicly available.
Multilingual pretrained language models (MPLMs) exhibit multilinguality and are well suited for transfer across languages. Most MPLMs are trained in an unsupervised fashion and the relationship between their objective and multilinguality is unclear.
More specifically, the question whether MPLM representations are language-agnostic or they simply interleave well with learned task prediction heads arises. In this work, we locate language-specific information in MPLMs and identify its dimensionality and the layers where this information occurs. We show that language-specific information is scattered across many dimensions, which can be projected into a linear subspace. Our study contributes to a better understanding of MPLM representations, going beyond treating them as unanalyzable blobs of information.
This paper describes the submissions of the NiuTrans Team to the WNGT 2020 Efficiency Shared Task. We focus on the efficient implementation of deep Transformer models cite{wang-etal-2019-learning, li-etal-2019-niutrans} using NiuTensor (https://githu
b.com/NiuTrans/NiuTensor), a flexible toolkit for NLP tasks. We explored the combination of deep encoder and shallow decoder in Transformer models via model compression and knowledge distillation. The neural machine translation decoding also benefits from FP16 inference, attention caching, dynamic batching, and batch pruning. Our systems achieve promising results in both translation quality and efficiency, e.g., our fastest system can translate more than 40,000 tokens per second with an RTX 2080 Ti while maintaining 42.9 BLEU on textit{newstest2018}. The code, models, and docker images are available at NiuTrans.NMT (https://github.com/NiuTrans/NiuTrans.NMT).
This paper describes the NiuTrans system for the WMT21 translation efficiency task (http://statmt.org/wmt21/efficiency-task.html). Following last years work, we explore various techniques to improve efficiency while maintaining translation quality. W
e investigate the combinations of lightweight Transformer architectures and knowledge distillation strategies. Also, we improve the translation efficiency with graph optimization, low precision, dynamic batching, and parallel pre/post-processing. Our system can translate 247,000 words per second on an NVIDIA A100, being 3$times$ faster than last years system. Our system is the fastest and has the lowest memory consumption on the GPU-throughput track. The code, model, and pipeline will be available at NiuTrans.NMT (https://github.com/NiuTrans/NiuTrans.NMT).
