Embeddings are generic representations that are useful for many NLP tasks. In this paper, we introduce DENSIFIER, a method that learns an orthogonal transformation of the embedding space that focuses the information relevant for a task in an ultradense subspace of a dimensionality that is smaller by a factor of 100 than the original space. We show that ultradense embeddings generated by DENSIFIER reach state of the art on a lexicon creation task in which words are annotated with three types of lexical information - sentiment, concreteness and frequency. On the SemEval2015 10B sentiment analysis task we show that no information is lost when the ultradense subspace is used, but training is an order of magnitude more efficient due to the compactness of the ultradense space.
Word embeddings are useful for a wide variety of tasks, but they lack interpretability. By rotating word spaces, interpretable dimensions can be identified while preserving the information contained in the embeddings without any loss. In this work, we investigate three methods for making word spaces interpretable by rotation: Densifier (Rothe et al., 2016), linear SVMs and DensRay, a new method we propose. In contrast to Densifier, DensRay can be computed in closed form, is hyperparameter-free and thus more robust than Densifier. We evaluate the three methods on lexicon induction and set-based…
Neural networks have excelled at many NLP tasks, but there remain open questions about the performance of pretrained distributed word representations and their interaction with weight initialization and other hyperparameters. We address these questions empirically using attention-based sequence-to-sequence models for natural language inference (NLI). Specifically, we compare three types of embeddings: random, pretrained (GloVe, word2vec), and retrofitted (pretrained plus WordNet information). We show that pretrained embeddings outperform both random and retrofitted ones in a large NLI corpus. …
Nanyang Technological University（南洋理工大学）；Institute of Computing Technology, Chinese Academy of Sciences（中国科学院计算技术研究所）
A sentence may express sentiments on multiple aspects. When these aspects are associated with different sentiment polarities, a model's accuracy is often adversely affected. We observe that multiple aspects in such hard sentences are mostly expressed through multiple clauses, or formally known as elementary discourse units (EDUs), and one EDU tends to express a single aspect with unitary sentiment towards that aspect. In this paper, we propose to consider EDU boundaries in sentence modeling, with attentions at both word and EDU levels. Specifically, we highlight sentiment-bearing words in EDU …
While deep learning models are making fast progress on the task of Natural Language Inference, recent studies have also shown that these models achieve high accuracy by exploiting several dataset biases, and without deep understanding of the language semantics. Using contradiction-word bias and word-overlapping bias as our two bias examples, this paper explores both data-level and model-level debiasing methods to robustify models against lexical dataset biases. First, we debias the dataset through data augmentation and enhancement, but show that the model bias cannot be fully removed via this …
Stanford University（斯坦福大学）；University of Toronto（多伦多大学）
A notable property of word embeddings is that word relationships can exist as linear substructures in the embedding space. For example, $\textit{gender}$ corresponds to $\vec{\textit{woman}} - \vec{\textit{man}}$ and $\vec{\textit{queen}} - \vec{\textit{king}}$. This, in turn, allows word analogies to be solved arithmetically: $\vec{\textit{king}} - \vec{\textit{man}} + \vec{\textit{woman}} \approx \vec{\textit{queen}}$. This property is notable because it suggests that models trained on word embeddings can easily learn such relationships as geometric translations. However, there is no evidenc…
We propose a simple and robust non-parameterized approach for building sentence representations. Inspired by the Gram-Schmidt Process in geometric theory, we build an orthogonal basis of the subspace spanned by a word and its surrounding context in a sentence. We model the semantic meaning of a word in a sentence based on two aspects. One is its relatedness to the word vector subspace already spanned by its contextual words. The other is the word's novel semantic meaning which shall be introduced as a new basis vector perpendicular to this existing subspace. Following this motivation, we devel…
National Research University Higher School of Economics（国家研究型高等经济大学）
Consider a continuous word embedding model. Usually, the cosines between word vectors are used as a measure of similarity of words. These cosines do not change under orthogonal transformations of the embedding space. We demonstrate that, using some canonical orthogonal transformations from SVD, it is possible both to increase the meaning of some components and to make the components more stable under re-learning. We study the interpretability of components for publicly available models for the Russian language (RusVectores, fastText, RDT).
This paper explores linear methods for combining several word embedding models into an ensemble. We construct the combined models using an iterative method based on either ordinary least squares regression or the solution to the orthogonal Procrustes problem. We evaluate the proposed approaches on Estonian---a morphologically complex language, for which the available corpora for training word embeddings are relatively small. We compare both combined models with each other and with the input word embedding models using synonym and analogy tests. The results show that while using the ordinary le…
University of West Bohemia（西波希米亚大学）；NTIS – New Technologies for the Information Society（NTIS——信息社会新技术中心）
In this paper, we describe our method for detection of lexical semantic change (i.e., word sense changes over time) for the DIACR-Ita shared task, where we ranked $1^{st}$. We examine semantic differences between specific words in two Italian corpora, chosen from different time periods. Our method is fully unsupervised and language independent. It consists of preparing a semantic vector space for each corpus, earlier and later. Then we compute a linear transformation between earlier and later spaces, using CCA and Orthogonal Transformation. Finally, we measure the cosines between the transform…
We experiment with new methods for learning how related words are positioned relative to each other in word embedding spaces. Previous approaches learned constant vector offsets: vectors that point from source tokens to target tokens with an assumption that these offsets were parallel to each other. We show that the offsets between related tokens are closer to orthogonal than parallel, and that they have low cosine similarities. We proceed by making a different assumption; target tokens are linearly separable from source and un-labeled tokens. We show that a max-margin hyperplane can separate …
Universidade de Vigo（维戈大学）；Cardiff University（卡迪夫大学）
Word embeddings have become a standard resource in the toolset of any Natural Language Processing practitioner. While monolingual word embeddings encode information about words in the context of a particular language, cross-lingual embeddings define a multilingual space where word embeddings from two or more languages are integrated together. Current state-of-the-art approaches learn these embeddings by aligning two disjoint monolingual vector spaces through an orthogonal transformation which preserves the structure of the monolingual counterparts. In this work, we propose to apply an addition…
Language representations are known to carry stereotypical biases and, as a result, lead to biased predictions in downstream tasks. While existing methods are effective at mitigating biases by linear projection, such methods are too aggressive: they not only remove bias, but also erase valuable information from word embeddings. We develop new measures for evaluating specific information retention that demonstrate the tradeoff between bias removal and information retention. To address this challenge, we propose OSCaR (Orthogonal Subspace Correction and Rectification), a bias-mitigating method th…
Usually bilingual word vectors are trained "online". Mikolov et al. showed they can also be found "offline", whereby two pre-trained embeddings are aligned with a linear transformation, using dictionaries compiled from expert knowledge. In this work, we prove that the linear transformation between two spaces should be orthogonal. This transformation can be obtained using the singular value decomposition. We introduce a novel "inverted softmax" for identifying translation pairs, with which we improve the precision @1 of Mikolov's original mapping from 34% to 43%, when translating a test set com…
The George Washington University（乔治·华盛顿大学）；AWS, Amazon AI（亚马逊云服务，亚马逊人工智能（AWS, Amazon AI））
Cross-lingual word vectors are typically obtained by fitting an orthogonal matrix that maps the entries of a bilingual dictionary from a source to a target vector space. Word vectors, however, are most commonly used for sentence or document-level representations that are calculated as the weighted average of word embeddings. In this paper, we propose an alternative to word-level mapping that better reflects sentence-level cross-lingual similarity. We incorporate context in the transformation matrix by directly mapping the averaged embeddings of aligned sentences in a parallel corpus. We also i…
Mila – Québec AI Institute（魁北克人工智能研究所Mila）；McGill University（麦吉尔大学）；Bank of Montréal（蒙特利尔银行）
Word embeddings are trained to predict word cooccurrence statistics, which leads them to possess different lexical properties (syntactic, semantic, etc.) depending on the notion of context defined at training time. These properties manifest when querying the embedding space for the most similar vectors, and when used at the input layer of deep neural networks trained to solve downstream NLP problems. Meta-embeddings combine multiple sets of differently trained word embeddings, and have been shown to successfully improve intrinsic and extrinsic performance over equivalent models which use just …
Tokyo Metropolitan University（东京都立大学）；University of Liverpool（利物浦大学）；Amazon（亚马逊）
Word embeddings trained on large corpora have shown to encode high levels of unfair discriminatory gender, racial, religious and ethnic biases. In contrast, human-written dictionaries describe the meanings of words in a concise, objective and an unbiased manner. We propose a method for debiasing pre-trained word embeddings using dictionaries, without requiring access to the original training resources or any knowledge regarding the word embedding algorithms used. Unlike prior work, our proposed method does not require the types of biases to be pre-defined in the form of word lists, and learns …
University of Southern California（南加州大学）；University of California, Berkeley（加州大学伯克利分校）
Although embedded vector representations of words offer impressive performance on many natural language processing (NLP) applications, the information of ordered input sequences is lost to some extent if only context-based samples are used in the training. For further performance improvement, two new post-processing techniques, called post-processing via variance normalization (PVN) and post-processing via dynamic embedding (PDE), are proposed in this work. The PVN method normalizes the variance of principal components of word vectors while the PDE method learns orthogonal latent variables fro…
With the recent success of pre-trained models in NLP, a significant focus was put on interpreting their representations. One of the most prominent approaches is structural probing (Hewitt and Manning, 2019), where a linear projection of word embeddings is performed in order to approximate the topology of dependency structures. In this work, we introduce a new type of structural probing, where the linear projection is decomposed into 1. isomorphic space rotation; 2. linear scaling that identifies and scales the most relevant dimensions. In addition to syntactic dependency, we evaluate our metho…
Negation Scope Resolution is an extensively researched problem, which is used to locate the words affected by a negation cue in a sentence. Recent works have shown that simply finetuning transformer-based architectures yield state-of-the-art results on this task. In this work, we look at Negation Scope Resolution as a Cloze-Style task, with the sentence as the Context and the cue words as the Query. We also introduce a novel Cloze-Style Attention mechanism called Orthogonal Attention, which is inspired by Self Attention. First, we propose a framework for developing Orthogonal Attention variant…
University of Maryland, College Park（马里兰大学帕克分校）；Massachusetts Institute of Technology（麻省理工学院）
Cross-lingual word embeddings (CLWE) underlie many multilingual natural language processing systems, often through orthogonal transformations of pre-trained monolingual embeddings. However, orthogonal mapping only works on language pairs whose embeddings are naturally isomorphic. For non-isomorphic pairs, our method (Iterative Normalization) transforms monolingual embeddings to make orthogonal alignment easier by simultaneously enforcing that (1) individual word vectors are unit length, and (2) each language's average vector is zero. Iterative Normalization consistently improves word translati…
We propose a geometric framework for learning meta-embeddings of words from different embedding sources. Our framework transforms the embeddings into a common latent space, where, for example, simple averaging of different embeddings (of a given word) is more amenable. The proposed latent space arises from two particular geometric transformations - the orthogonal rotations and the Mahalanobis metric scaling. Empirical results on several word similarity and word analogy benchmarks illustrate the efficacy of the proposed framework.
University of California Los Angeles（加州大学洛杉矶分校）；University of Southern California（南加利福尼亚大学）；University of Pennsylvania（宾夕法尼亚大学）
Contextualized word embedding models, such as ELMo, generate meaningful representations of words and their context. These models have been shown to have a great impact on downstream applications. However, in many cases, the contextualized embedding of a word changes drastically when the context is paraphrased. As a result, the downstream model is not robust to paraphrasing and other linguistic variations. To enhance the stability of contextualized word embedding models, we propose an approach to retrofitting contextualized embedding models with paraphrase contexts. Our method learns an orthogo…
University of Virginia（弗吉尼亚大学）；Salesforce Research（赛富时研究院）
Word embeddings derived from human-generated corpora inherit strong gender bias which can be further amplified by downstream models. Some commonly adopted debiasing approaches, including the seminal Hard Debias algorithm, apply post-processing procedures that project pre-trained word embeddings into a subspace orthogonal to an inferred gender subspace. We discover that semantic-agnostic corpus regularities such as word frequency captured by the word embeddings negatively impact the performance of these algorithms. We propose a simple but effective technique, Double Hard Debias, which purifies …
Cross-lingual word embeddings aim to bridge the gap between high-resource and low-resource languages by allowing to learn multilingual word representations even without using any direct bilingual signal. The lion's share of the methods are projection-based approaches that map pre-trained embeddings into a shared latent space. These methods are mostly based on the orthogonal transformation, which assumes language vector spaces to be isomorphic. However, this criterion does not necessarily hold, especially for morphologically-rich languages. In this paper, we propose a self-supervised method to …
State-of-the-art contextual embeddings are obtained from large language models available only for a few languages. For others, we need to learn representations using a multilingual model. There is an ongoing debate on whether multilingual embeddings can be aligned in a space shared across many languages. The novel Orthogonal Structural Probe (Limisiewicz and Mareček, 2021) allows us to answer this question for specific linguistic features and learn a projection based only on mono-lingual annotated datasets. We evaluate syntactic (UD) and lexical (WordNet) structural information encoded inmBERT…
BioISI—Instituto de Biosistemas e Ciências Integrativas（BioISI——生物系统与综合科学研究所）；Faculdade de Ciências, Universidade de Lisboa（里斯本大学理学院）
Neural network-based language models deal with data sparsity problems by mapping the large discrete space of words into a smaller continuous space of real-valued vectors. By learning distributed vector representations for words, each training sample informs the neural network model about a combinatorial number of other patterns. In this paper, we exploit the sparsity in natural language even further by encoding each unique input word using a fixed sparse random representation. These sparse codes are then projected onto a smaller embedding space which allows for the encoding of word occurrences…
Typically, a linearly orthogonal transformation mapping is learned by aligning static type-level embeddings to build a shared semantic space. In view of the analysis that contextual embeddings contain richer semantic features, we investigate a context-aware and dictionary-free mapping approach by leveraging parallel corpora. We illustrate that our contextual embedding space mapping significantly outperforms previous multilingual word embedding methods on the bilingual dictionary induction (BDI) task by providing a higher degree of isomorphism. To improve the quality of mapping, we also explore…
The present paper explores a novel variant of Random Indexing (RI) based representations for encoding language data with a view to using them in a dynamic scenario where events are happening in a continuous fashion. As the size of the representations in the general method of onehot encoding grows linearly with the size of the vocabulary, they become non-scalable for online purposes with high volumes of dynamic data. On the other hand, existing pre-trained embedding models are not suitable for detecting happenings of new events due to the dynamic nature of the text data. The present work addres…
Facebook AI Research（脸书人工智能研究院）；University of Cambridge（剑桥大学）；Statistical Laboratory（统计实验室）
We consider the task of aligning two sets of points in high dimension, which has many applications in natural language processing and computer vision. As an example, it was recently shown that it is possible to infer a bilingual lexicon, without supervised data, by aligning word embeddings trained on monolingual data. These recent advances are based on adversarial training to learn the mapping between the two embeddings. In this paper, we propose to use an alternative formulation, based on the joint estimation of an orthogonal matrix and a permutation matrix. While this problem is not convex, …
The study of the applicability of the BERTScore metric was conducted to translation quality assessment at the sentence level for English -> Russian direction. Experiments were performed with a pre-trained Multilingual BERT as well as with a pair of Monolingual BERT models. To align monolingual embeddings, an orthogonal transformation based on anchor tokens was used. It was demonstrated that such transformation helps to prevent mismatching issue and shown that this approach gives better results than using embeddings of the Multilingual model. To improve the token matching process it is proposed…
