Large-scale retrieval is to recall relevant documents from a huge collection given a query. It relies on representation learning to embed documents and queries into a common semantic encoding space. According to the encoding space, recent retrieval methods based on pre-trained language models (PLM) can be coarsely categorized into either dense-vector or lexicon-based paradigms. These two paradigms unveil the PLMs' representation capability in different granularities, i.e., global sequence-level compression and local word-level contexts, respectively. Inspired by their complementary global-local contextualization and distinct representing views, we propose a new learning framework, Unifier, which unifies dense-vector and lexicon-based retrieval in one model with a dual-representing capability. Experiments on passage retrieval benchmarks verify its effectiveness in both paradigms. A uni-retrieval scheme is further presented with even better retrieval quality. We lastly evaluate the model on BEIR benchmark to verify its transferability.

Retrieval models based on dense representations in semantic space have become an indispensable branch for first-stage retrieval. These retrievers benefit from surging advances in representation learning towards compressive global sequence-level embeddings. However, they are prone to overlook local salient phrases and entity mentions in texts, which usually play pivot roles in first-stage retrieval. To mitigate this weakness, we propose to make a dense retriever align a well-performing lexicon-aware representation model. The alignment is achieved by weakened knowledge distillations to enlighten the retriever via two aspects – 1) a lexicon-augmented contrastive objective to challenge the dense encoder and 2) a pair-wise rank-consistent regularization to make the dense model’s behavior incline to the other. We evaluate our model on three public benchmarks, which shows that with a comparable lexicon-aware retriever as the teacher, our proposed dense one can bring consistent and significant improvements, and even outdo its teacher. In addition, we show our lexicon-aware distillation strategies are compatible with the standard ranker distillation, which can further lift state-of-the-art performance.1

Event correlation reasoning infers whether a natural language paragraph containing multiple events conforms to human common sense. For example, “Andrew was very drowsy, so he took a long nap, and now he is very alert” is sound and reasonable. In contrast, “Andrew was very drowsy, so he stayed up a long time, now he is very alert” does not comply with human common sense. Such reasoning capability is essential for many downstream tasks, such as script reasoning, abductive reasoning, narrative incoherence, story cloze test, etc. However, conducting event correlation reasoning is challenging due to a lack of large amounts of diverse event-based knowledge and difficulty in capturing correlation among multiple events. In this paper, we propose EventBERT, a pre-trained model to encapsulate eventuality knowledge from unlabeled text. Specifically, we collect a large volume of training examples by identifying natural language paragraphs that describe multiple correlated events and further extracting event spans in an unsupervised manner. We then propose three novel event- and correlation-based learning objectives to pre-train an event correlation model on our created training corpus. Experimental results show EventBERT outperforms strong baselines on four downstream tasks, and achieves state-of-the-art results on most of them. Moreover, it outperforms existing pre-trained models by a large margin, e.g., 6.5 ∼ 23%, in zero-shot learning of these tasks.

Language scaling aims to deploy Natural Language Processing (NLP) applications economically across many countries/regions with different languages. Language scaling has been heavily invested by industry since many parties want to deploy their applications/services to global markets. At the same time, scaling out NLP applications to various languages, essentially a data science problem, remains a grand challenge due to the huge differences in the morphology, syntaxes, and pragmatics among different languages. We present a comprehensive survey and tutorial on language scaling. We start with a clear problem description for language scaling and an intuitive discussion on the overall challenges. Then, we outline two major categories of approaches to language scaling, namely, model transfer and data transfer. We present a taxonomy to summarize various methods in literature. A large part of the tutorial is organized to address various types of NLP applications. Finally, we discuss several important challenges in this area and future directions.

Entity candidate retrieval plays a critical role in cross-lingual entity linking (XEL). In XEL, entity candidate retrieval needs to retrieve a list of plausible candidate entities from a large knowledge graph in a target language given a piece of text in a sentence or question, namely a mention, in a source language. Existing works mainly fall into two categories: lexicon-based and semantic-based approaches. The lexicon-based approach usually creates cross-lingual and mention-entity lexicons, which is effective but relies heavily on bilingual resources (e.g. inter-language links in Wikipedia). The semantic-based approach maps mentions and entities in different languages to a unified embedding space, which reduces dependence on large-scale bilingual dictionaries. However, its effectiveness is limited by the representation capacity of fixed-length vectors. In this paper, we propose a pivot-based approach which inherits the advantages of the aforementioned two approaches while avoiding their limitations. It takes an intermediary set of plausible target-language mentions as pivots to bridge the two types of gaps: cross-lingual gap and mention-entity gap. Specifically, it first converts mentions in the source language into an intermediary set of plausible mentions in the target language by cross-lingual semantic retrieval and a selective mechanism, and then retrieves candidate entities based on the generated mentions by lexical retrieval. The proposed approach only relies on a small bilingual word dictionary, and fully exploits the benefits of both lexical and semantic matching. Experimental results on two challenging cross-lingual entity linking datasets spanning over 11 languages show that the pivot-based approach outperforms both the lexicon-based and semantic-based approach by a large margin.

Procedural text describes dynamic state changes during a step-by-step natural process (e.g., photosynthesis). In this work, we focus on the task of procedural text understanding, which aims to comprehend such documents and track entities’ states and locations during a process. Although recent approaches have achieved substantial progress, their results are far behind human performance. Two challenges, the difficulty of commonsense reasoning and data insufficiency, still remain unsolved, which require the incorporation of external knowledge bases. Previous works on external knowledge injection usually rely on noisy web mining tools and heuristic rules with limited applicable scenarios. In this paper, we propose a novel KnOwledge-Aware proceduraL text understAnding (KoaLa) model, which effectively leverages multiple forms of external knowledge in this task. Specifically, we retrieve informative knowledge triples from ConceptNet and perform knowledge-aware reasoning while tracking the entities. Besides, we employ a multi-stage training schema which fine-tunes the BERT model over unlabeled data collected from Wikipedia before further fine-tuning it on the final model. Experimental results on two procedural text datasets, ProPara and Recipes, verify the effectiveness of the proposed methods, in which our model achieves state-of-the-art performance in comparison to various baselines.1

Recent studies on open-domain question answering have achieved prominent performance improvement using pre-trained language models such as BERT. State-of-the-art approaches typically follow the "retrieve and read" pipeline and employ BERT-based reranker to filter retrieved documents before feeding them into the reader module. The BERT retriever takes as input the concatenation of question and each retrieved document. Despite the success of these approaches in terms of QA accuracy, due to the concatenation, they can barely handle high-throughput of incoming questions each with a large collection of retrieved documents. To address the efficiency problem, we propose DC-BERT, a decoupled contextual encoding framework that has dual BERT models: an online BERT which encodes the question only once, and an offline BERT which pre-encodes all the documents and caches their encodings. On SQuAD Open and Natural Questions Open datasets, DC-BERT achieves 10x speedup on document retrieval, while retaining most (about 98%) of the QA performance compared to state-of-the-art approaches for open-domain question answering.

Learning to rank, which learns the ranking function from training data, has become an emerging research area in information retrieval and machine learning. Most existing work on learning to rank assumes that the training data is clean, which is not always true, however. The ambiguity of query intent, the lack of domain knowledge, and the vague definition of relevance levels all make it difficult for common annotators to give reliable relevance labels to some documents. As a result, the relevance labels in the training data of learning to rank usually contain noise. If we ignore this fact, the performance of learning-to-rank algorithms will be damaged. In this article, we propose considering the labeling noise in the process of learning to rank and using a two-step approach to extend existing algorithms to handle noisy training data. In the first step, we estimate the degree of labeling noise for a training document. To this end, we assume that the majority of the relevance labels in the training data are reliable and we use a graphical model to describe the generative process of a training query, the feature vectors of its associated documents, and the relevance labels of these documents. The parameters in the graphical model are learned by means of maximum likelihood estimation. Then the conditional probability of the relevance label given the feature vector of a document is computed. If the probability is large, we regard the degree of labeling noise for this document as small; otherwise, we regard the degree as large. In the second step, we extend existing learning-to-rank algorithms by incorporating the estimated degree of labeling noise into their loss functions. Specifically, we give larger weights to those training documents with smaller degrees of labeling noise and smaller weights to those with larger degrees of labeling noise. As examples, we demonstrate the extensions for McRank, RankSVM, RankBoost, and RankNet. Empirical results on benchmark datasets show that the proposed approach can effectively distinguish noisy documents from clean ones, and the extended learning-to-rank algorithms can achieve better performances than baselines.

When applying learning to rank algorithms in real search applications, noise in human labeled training data becomes an inevitable problem which will affect the performance of the algorithms. Previous work mainly focused on studying how noise affects ranking algorithms and how to design robust ranking algorithms. In our work, we investigate what inherent characteristics make training data robust to label noise. The motivation of our work comes from an interesting observation that a same ranking algorithm may show very different sensitivities to label noise over different data sets. We thus investigate the underlying reason for this observation based on two typical kinds of learning to rank algorithms (i.e.~pairwise and listwise methods) and three different public data sets (i.e.~OHSUMED, TD2003 and MSLR-WEB10K). We find that when label noise increases in training data, it is the \emph{document pair noise ratio} (i.e.~\emph{pNoise}) rather than \emph{document noise ratio} (i.e.~\emph{dNoise}) that can well explain the performance degradation of a ranking algorithm.

We consider the problem of extracting, in a domain-centric fashion, a given set of attributes from a large number of semi-structured websites. Previous approaches [7, 5] to solve this problem are based on page level inference. We propose a distinct new approach that directly chooses attribute extractors for a site using a scoring mechanism that is designed at the domain level via simple classification methods using a training set from a small number of sites. To keep the number of candidate extractors in each site manageably small we use two observations that hold in most domains: (a) imprecise annotators can be used to identify a small set of candidate extractors for a few attributes (anchors); and (b) non-anchor attributes lie in close proximity to the anchor attributes. Experiments on three domains (Events, Books and Restaurants) show that our approach is very effective in spite of its simplicity.

Many ranking models have been proposed in information retrieval, and recently machine learning techniques have also been applied to ranking model construction. Most of the existing methods do not take into consideration the fact that significant differences exist between queries, and only resort to a single function in ranking of documents. In this paper, we argue that it is necessary to employ different ranking models for different queries and onduct what we call query-dependent ranking. As the first such attempt, we propose a K-Nearest Neighbor (KNN) method for query-dependent ranking. We first consider an online method which creates a ranking model for a given query by using the labeled neighbors of the query in the query feature space and then rank the documents with respect to the query using the created model. Next, we give two offline approximations of the method, which create the ranking models in advance to enhance the efficiency of ranking. And we prove a theory which indicates that the approximations are accurate in terms of difference in loss of prediction, if the learning algorithm used is stable with respect to minor changes in training examples. Our experimental results show that the proposed online and offline methods both outperform the baseline method of using a single ranking function.

Ranking is a very important topic in information retrieval. While algorithms for learning ranking models have been intensively studied, this is not the case for feature selection, despite of its importance. The reality is that many feature selection methods used in classification are directly applied to ranking. We argue that because of the striking differences between ranking and classification, it is better to develop different feature selection methods for ranking. To this end, we propose a new feature selection method in this paper. Specifically, for each feature we use its value to rank the training instances, and define the ranking accuracy in terms of a performance measure or a loss function as the importance of the feature. We also define the correlation between the ranking results of two features as the similarity between them. Based on the definitions, we formulate the feature selection issue as an optimization problem, for which it is to find the features with maximum total importance scores and minimum total similarity scores. We also demonstrate how to solve the optimization problem in an efficient way. We have tested the effectiveness of our feature selection method on two information retrieval datasets and with two ranking models. Experimental results show that our method can outperform traditional feature selection methods for the ranking task.