DOCSLIB.ORG

A Compositional and Interpretable Semantic Space

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
A Compositional and Interpretable Semantic Space A Compositional and Interpretable Semantic Space Alona Fyshe,1 Leila Wehbe,1 Partha Talukdar,2 Brian Murphy,3 and Tom Mitchell1 1 Machine Learning Department, Carnegie Mellon University, Pittsburgh, USA 2 Indian Institute of Science, Bangalore, India 3 Queen’s University Belfast, Belfast, Northern Ireland [email protected], [email protected], [email protected], [email protected], [email protected] Abstract like a black box - it is unclear what VSM dimen- sions represent (save for broad classes of corpus Vector Space Models (VSMs) of Semantics statistic types) and what the application of a com- are useful tools for exploring the semantics of position function to those dimensions entails. Neu- single words, and the composition of words ral network (NN) models are becoming increas- to make phrasal meaning. While many meth- ingly popular (Socher et al., 2012; Hashimoto et al., ods can estimate the meaning (i.e. vector) of a phrase, few do so in an interpretable way. 2014; Mikolov et al., 2013; Pennington et al., 2014), We introduce a new method (CNNSE) that al- and some model introspection has been attempted: lows word and phrase vectors to adapt to the Levy and Goldberg (2014) examined connections notion of composition. Our method learns a between layers, Mikolov et al. (2013) and Penning- VSM that is both tailored to support a chosen ton et al. (2014) explored how shifts in VSM space semantic composition operation, and whose encodes semantic relationships. Still, interpreting resulting features have an intuitive interpreta- NN VSM dimensions, or factors, remains elusive. tion. Interpretability allows for the exploration of phrasal semantics, which we leverage to an- This paper introduces a new method, Composi- alyze performance on a behavioral task. tional Non-negative Sparse Embedding (CNNSE). In contrast to many other VSMs, our method learns an interpretable VSM that is tailored to suit the se- 1 Introduction mantic composition function. Such interpretability Vector Space Models (VSMs) are models of word allows for deeper exploration of semantic composi- semantics typically built with word usage statistics tion than previously possible. We will begin with an derived from corpora. VSMs have been shown to overview of the CNNSE algorithm, and follow with closely match human judgements of semantics (for empirical results which show that CNNSE produces: an overview see Sahlgren (2006), Chapter 5), and 1. more interpretable dimensions than the typical can be used to study semantic composition (Mitchell VSM, and Lapata, 2010; Baroni and Zamparelli, 2010; 2. composed representations that outperform pre- Socher et al., 2012; Turney, 2012). vious methods on a phrase similarity task. Composition has been explored with different types of composition functions (Mitchell and La- Compared to methods that do not consider composi- pata, 2010; Mikolov et al., 2013; Dinu et al., tion when learning embeddings, CNNSE produces: 2013) including higher order functions (such as ma- trices) (Baroni and Zamparelli, 2010), and some 1. better approximations of phrasal semantics, have considered which corpus-derived information 2. phrasal representations with dimensions that is most useful for semantic composition (Turney, more closely match phrase meaning. 2012; Fyshe et al., 2013). Still, many VSMs act 32 Human Language Technologies: The 2015 Annual Conference of the North American Chapter of the ACL, pages 32–41, Denver, Colorado, May 31 – June 5, 2015. c 2015 Association for Computational Linguistics 2 Method mensional representation for w words using the c- w c Typically, word usage statistics used to create a dimensional corpus statistics in a matrix X R × . w ∈` The solution is two matrices: A R × that is VSM form a sparse matrix with many columns, too ∈ unwieldy to be practical. Thus, most models use sparse, non-negative, and represents word semantics ` c in an `-dimensional latent space, and D R × : some form of dimensionality reduction to compress ∈ the full matrix. For example, Latent Semantic Anal- the encoding of corpus statistics in the latent space. ysis (LSA) (Deerwester et al., 1990) uses Singular NNSE minimizes the following objective: w Value Decomposition (SVD) to create a compact 1 argmin X A D 2 + λ A VSM. SVD often produces matrices where, for the i,: i,: 1 i,: 1 A,D 2 − × vast majority of the dimensions, it is difficult to in- Xi=1 (1) terpret what a high or low score entails for the se- st: D DT 1, 1 i ` (2) mantics of a given word. In addition, the SVD fac- i,: i,: ≤ ∀ ≤ ≤ torization does not take into account the phrasal re- Ai,j 0, 1 i w, 1 j ` (3) lationships between the input words. ≥ ≤ ≤ ≤ ≤ where Ai,j indicates the entry at the ith row and jth 2.1 Non-negative Sparse Embeddings column of matrix A, and Ai,: indicates the ith row Our method is inspired by Non-negative Sparse Em- of the matrix. The L1 constraint encourages sparsity beddings (NNSEs) (Murphy et al., 2012). NNSE in A; λ1 is a hyperparameter. Equation 2 constrains promotes interpretability by including sparsity and D to eliminate solutions where the elements of A non-negativity constraints into a matrix factoriza- are made arbitrarily small by making the norm of D tion algorithm. The result is a VSM with extremely arbitrarily large. Equation 3 ensures that A is non- coherent dimensions, as quantified by a behavioral negative. Together, A and D factor the original cor- task (Murphy et al., 2012). The output of NNSE pus statistics matrix X to minimize reconstruction is a matrix with rows corresponding to words and error. One may tune ` and λ1 to vary the sparsity of columns corresponding to latent dimensions. the final solution. To interpret a particular latent dimension, we can Murphy et al. (2012) solved this system of con- examine the words with the highest numerical val- straints using the Online Dictionary Learning algo- ues in that dimension (i.e. identify rows with the rithm described in Mairal et al. (2010). Though highest values for a particular column). Though the Equations 1-3 represent a non-convex system, when representations in Table 1 were created with our new solving for A with D fixed (and vice versa) the loss method, CNNSE, we will use them to illustrate the function is convex. Mairal et al. break the prob- interpretability of both NNSE and CNNSE, as the lem into two alternating optimization steps (solv- form of the learned representations is similar. One ing for A and D) and find the system converges of the dimensions in Table 1 has top scoring words to a stationary solution. The solution for A is guidance, advice and assistance - words related to found with a LARS implementation for lasso regres- help and support. We will refer to these word list sion (Efron et al., 2004); D is found via gradient de- summaries as the dimension’s interpretable sum- scent. Though the final solution may not be globally marization. To interpret the meaning of a particu- optimal, this method is capable of handling large lar word, we can select its highest scoring dimen- amounts of data and has been shown to produce use- sions (i.e. choose columns with maximum values ful solutions in practice (Mairal et al., 2010; Murphy for a particular row). For example, the interpretable et al., 2012). summarizations for the top scoring dimensions of the word military include both positions in the mil- 2.2 Compositional NNSE itary (e.g. commandos), and military groups (e.g. We add an additional constraint to the NNSE loss paramilitary). More examples in Supplementary function that allows us to learn a latent representa- Material (http://www.cs.cmu.edu/˜fmri/ tion that respects the notion of semantic composi- papers/naacl2015/). tion. As we will see, this change to the loss function NNSE is an algorithm which seeks a lower di- has a huge effect on the learned latent space. Just as 2 33 Table 1: CNNSE interpretable summarizations for the top 3 dimensions of an adjective, noun and adjective- noun phrase. military aid military aid (observed) servicemen, commandos, guidance, advice, assistance servicemen, commandos, military intelligence military intelligence guerrilla, paramilitary, anti-terrorist mentoring, tutoring, internships guidance, advice, assistance conglomerate, giants, conglomerates award, awards, honors compliments, congratulations, replies the L1 regularizer can have a large impact on spar- This choice of f requires that we simultaneously op- sity, our composition constraint represents a consid- timize for A, D, α and β. However, α and β are sim- erable change in composition compatibility. ply constant scaling factors for the vectors in A cor- Consider a phrase p made up of words i and j. In responding to adjectives and nouns. For adjective- the most general setting, the following composition noun composition, the optimization of α and β can constraint could be applied to the rows of matrix A be absorbed by the optimization of A. For models corresponding to p, i and j: that include noun-noun composition, if α and β are assumed to be absorbed by the optimization of A, A(p,:) = f(A(i,:),A(j,:)) (4) this is equivalent to setting α = β. where f is some composition function. The com- We can further simplify the loss function by con- position function constrains the space of learned la- structing a matrix B that imposes the composition w ` tent representations A R × to be those solutions ∈ by addition constraint. B is constructed so that for that are compatible with the composition function each phrase p = (i, j): B = 1, B = α, (p,p) (p,i) − defined by f. Incorporating f into Equation 1 we and B = β. For our models, we use α = β = (p,j) − have: 0.5, which serves to average the single word repre- w 1 sentations.
Load more

Recommended publications
  • Arguments For: Semantic Folding and Hierarchical Temporal Memory
    [email protected] Mariahilferstrasse 4 1070 Vienna, Austria http://www.cortical.io ARGUMENTS FOR: SEMANTIC FOLDING AND HIERARCHICAL TEMPORAL MEMORY Replies to the most common Semantic Folding criticisms Francisco Webber, September 2016 INTRODUCTION 2 MOTIVATION FOR SEMANTIC FOLDING 3 SOLVING ‘HARD’ PROBLEMS 3 ON STATISTICAL MODELING 4 FREQUENT CRITICISMS OF SEMANTIC FOLDING 4 “SEMANTIC FOLDING IS JUST WORD EMBEDDING” 4 “WHY NOT USE THE OPEN SOURCED WORD2VEC?” 6 “SEMANTIC FOLDING HAS NO COMPARATIVE EVALUATION” 7 “SEMANTIC FOLDING HAS ONLY LOGICAL/PHILOSOPHICAL ARGUMENTS” 7 FREQUENT CRITICISMS OF HTM THEORY 8 “JEFF HAWKINS’ WORK HAS BEEN MOSTLY PHILOSOPHICAL AND LESS TECHNICAL” 8 “THERE ARE NO MATHEMATICALLY SOUND PROOFS OR VALIDATIONS OF THE HTM ASSERTIONS” 8 “THERE IS NO REAL-WORLD SUCCESS” 8 “THERE IS NO COMPARISON WITH MORE POPULAR ALGORITHMS OF DEEP LEARNING” 8 “GOOGLE/DEEP MIND PRESENT THEIR PERFORMANCE METRICS, WHY NOT NUMENTA” 9 FMRI SUPPORT FOR SEMANTIC FOLDING THEORY 9 REALITY CHECK: BUSINESS APPLICABILITY 9 GLOBAL INDUSTRY UNDER PRESSURE 10 “WE HAVE TRIED EVERYTHING …” 10 REFERENCES 10 Arguments for Semantic Folding and Hierarchical Temporal Memory Theory Introduction During the last few years, the big promise of computer science, to solve any computable problem given enough data and a gold standard, has triggered a race for machine learning (ML) among tech communities. Sophisticated computational techniques have enabled the tackling of problems that have been considered unsolvable for decades. Face recognition, speech recognition, self- driving cars; it seems like a computational model could be created for any human task. Anthropology has taught us that a sufficiently complex technology is indistinguishable from magic by the non-expert, indigenous mind.
    [Show full text]
  • Embeddings in Natural Language Processing
    Embeddings in Natural Language Processing Theory and Advances in Vector Representation of Meaning Mohammad Taher Pilehvar Tehran Institute for Advanced Studies Jose Camacho-Collados Cardiff University SYNTHESIS LECTURESDRAFT ON HUMAN LANGUAGE TECHNOLOGIES M &C Morgan& cLaypool publishers ABSTRACT Embeddings have been one of the dominating buzzwords since the early 2010s for Natural Language Processing (NLP). Encoding information into a low-dimensional vector representation, which is easily integrable in modern machine learning algo- rithms, has played a central role in the development in NLP. Embedding techniques initially focused on words but the attention soon started to shift to other forms: from graph structures, such as knowledge bases, to other types of textual content, such as sentences and documents. This book provides a high level synthesis of the main embedding techniques in NLP, in the broad sense. The book starts by explaining conventional word vector space models and word embeddings (e.g., Word2Vec and GloVe) and then moves to other types of embeddings, such as word sense, sentence and document, and graph embeddings. We also provide an overview on the status of the recent development in contextualized representations (e.g., ELMo, BERT) and explain their potential in NLP. Throughout the book the reader can find both essential information for un- derstanding a certain topic from scratch, and an in-breadth overview of the most successful techniques developed in the literature. KEYWORDS Natural Language Processing, Embeddings, Semantics DRAFT iii Contents 1 Introduction ................................................. 1 1.1 Semantic representation . 3 1.2 One-hot representation . 4 1.3 Vector Space Models . 5 1.4 The Evolution Path of representations .
    [Show full text]
  • Sociolinguistic Properties of Word Embeddings Introduction
    Sociolinguistic Properties of Word Embeddings Arseniev-Koehler and Foster (Draft) Sociolinguistic Properties of Word Embeddings Alina Arseniev-Koehler and Jacob G. Foster UCLA Department of Sociology Introduction Just a single word can prime our thinking and initiate a stream of associations. Not only do words denote concepts or things-in-the-world; they can also be combined to convey complex ideas and evoke connotations and even feelings. The meanings of our words are also contextual — they vary across time, space, linguistic communities, interactions, and even specific sentences. As rich as words are to us, in their raw form they are merely a sequence of phonemes or characters. Computers typically encounter words in their written form. How can this sequence become meaningful to computers? Word embeddings are one approach to represent word meanings numerically. This representation enables computers to process language semantically as well as syntactically. Embeddings gained popularity because the “meaning” they captured corresponds to human meanings in unprecedented ways. In this chapter, we illustrate these surprising correspondences at length. Because word embeddings capture meaning so effectively, they are a key ingredient in a variety of downstream tasks involving natural language (Artetxe et al., 2017; Young et al., 2018). They are used ubiquitously in tasks like translating languages (Artetxe et al., 2017; Mikolov, Tomas, Quoc V. Le, and Ilya Sutskever, 2013) and parsing clinical notes (Ching et al., 2018). But embeddings are more than just tools for computers to represent words. Word embeddings can represent human meanings because they learn the meanings of words from human language use – such as news, books, crawling the web, or even television and movie scripts.
    [Show full text]
  • Semantic Structure and Interpretability of Word Embeddings
    1 Semantic Structure and Interpretability of Word Embeddings Lutfi¨ Kerem S¸enel, Student Member, IEEE, Ihsan˙ Utlu, Veysel Yucesoy,¨ Student Member, IEEE, Aykut Koc¸, Member, IEEE, and Tolga C¸ukur, Senior Member, IEEE, Abstract—Dense word embeddings, which encode meanings have been receiving increasing interest in NLP research. of words to low dimensional vector spaces, have become very Approaches that learn embeddings include neural network popular in natural language processing (NLP) research due to based predictive methods [2], [8] and count-based matrix- their state-of-the-art performances in many NLP tasks. Word embeddings are substantially successful in capturing semantic factorization methods [9]. Word embeddings brought about relations among words, so a meaningful semantic structure must significant performance improvements in many intrinsic NLP be present in the respective vector spaces. However, in many tasks such as analogy or semantic textual similarity tasks, cases, this semantic structure is broadly and heterogeneously as well as downstream NLP tasks such as part-of-speech distributed across the embedding dimensions making interpre- (POS) tagging [10], named entity recognition [11], word sense tation of dimensions a big challenge. In this study, we propose a statistical method to uncover the underlying latent semantic disambiguation [12], sentiment analysis [13] and cross-lingual structure in the dense word embeddings. To perform our analysis, studies [14]. we introduce a new dataset (SEMCAT) that contains more Although high levels of success have been reported in many than 6,500 words semantically grouped under 110 categories. NLP tasks using word embeddings, the individual embedding We further propose a method to quantify the interpretability dimensions are commonly considered to be uninterpretable of the word embeddings.
    [Show full text]
  • A Vector Semantics Approach to the Geoparsing Disambiguation Task for Texts in Spanish
    Kalpa Publications in Computing Volume 13, 2019, Pages 47{55 Proceedings of the 1st International Con- ference on Geospatial Information Sciences A Vector Semantics Approach to the Geoparsing Disambiguation Task for Texts in Spanish Filomeno Alc´antara1, Alejandro Molina2, and Victor Mu~niz3 1 Centro de Investigacion en Matem´aticas,Monterrey, Nuevo Le´on,M´exico [email protected] http://www.cimat.mx 2 CONACYT { Centro de Investigaci´onen Ciencias de Informaci´onGeoespacial, M´erida,Yucat´an, M´exico [email protected] http://mid.geoint.mx 3 Centro de Investigaci´onen Matem´aticas,Monterrey, Nuevo Le´on,M´exico victor [email protected] http://www.cimat.mx Abstract Nowadays, online news sources generate continuous streams of information that in- cludes references to real locations. Linking these locations to coordinates in a map usually requires two steps involving the named entity: extraction and disambiguation. In past years, efforts have been devoted mainly to the first task. Approaches to location disam- biguation include knowledge-based, map-based and data-driven methods. In this paper, we present a work in progress for location disambiguation in news documents that uses a vector-semantic representation learned from information sources that include events and geographic descriptions, in order to obtain a ranking for the possible locations. We will describe the proposed method and the results obtained so far, as well as ideas for future work. 1 Introduction Nowadays, online news sources generate continuous streams of information with global coverage that directly impact the daily activities of people in a wide variety of areas. Although social media can also generate real-time news content, newspapers remain a source of information that incorporates a refined version of relevant events.
    [Show full text]
  • The Discursive Construction of Migrant Otherness on Facebook: a Distributional Semantics Approach
    The discursive construction of migrant otherness on Facebook: A distributional semantics approach Victoria Yantseva∗ June 1, 2021 Abstract The goal of this work is to study the social construction of migrant categories and immigration discourse on Swedish Facebook in the last decade. I combine the insights from computational linguistics and distributional semantics approach with those from classical sociological theories in order to explore a corpus of more than 1M Facebook posts. This allows to compare the intended meanings of various linguistic labels denoting voluntary versus forced character of migration, as well as to distinguish the most salient themes that constitute the Facebook discourse. The study concludes that, although Facebook seems to have the highest potential in the promotion of toler- ance and support for migrants, its audience is nevertheless active in the discursive discrimination of those identified as \refugees" or \immigrants". The results of the study are then related to the technological design of new media and the overall social and political climate surrounding the Swedish immigration agenda. Keywords: word vectors, vector space models, word2vec, doc2vec, Facebook, social media, corpus linguistics, discourse analysis, migration 1 Introduction Language is an essential part of our everyday interactions in society and helps us make sense of the things we see, hear and feel. A widely accepted statement in social studies is that our reality is socially constructed, and language can quite naturally be seen as a necessary component in this process. The perceptions of human migration, for instance, can been named as one of such social constructs that are collectively created and manipulated.
    [Show full text]
  • Enhancing Domain Word Embedding Via Latent Semantic Imputation
    Enhancing Domain Word Embedding via Latent Semantic Imputation Shibo Yao Dantong Yu∗ Keli Xiao New Jersey Institute of Technology New Jersey Institute of Technology Stony Brook University Newark, United States Newark, United States Stony Brook, United States [email protected] [email protected] [email protected] ABSTRACT tasks. There are two main types of word embedding approaches We present a novel method named Latent Semantic Imputation (LSI) aiming at mapping words to real-valued vectors: matrix factoriza- to transfer external knowledge into semantic space for enhancing tion [7, 23, 29] and neural networks [4, 26]. The embedding matrix word embedding. The method integrates graph theory to extract is a reflection of the relative positions of words and phrases in the latent manifold structure of the entities in the affinity space and semantic spaces. Embedding vectors pre-trained on large corpora leverages non-negative least squares with standard simplex con- can serve as the embedding layer in neural network models and straints and power iteration method to derive spectral embeddings. may significantly benefit downstream language tasks because reli- It provides an effective and efficient approach to combining entity able external information is transferred from large corpora to those representations defined in different Euclidean spaces. Specifically, specific tasks. our approach generates and imputes reliable embedding vectors Word embedding techniques have achieved remarkable successes for low-frequency words in the semantic space and benefits down- in recent years. Nonetheless, there are still critical questions that are stream language tasks that depend on word embedding. We conduct not well answered in the literature.
    [Show full text]
  • Natural Language Processing (CSEP 517): Distributional Semantics
    Natural Language Processing (CSEP 517): Distributional Semantics Roy Schwartz c 2017 University of Washington [email protected] May 15, 2017 1 / 59 To-Do List I Read: (Jurafsky and Martin, 2016a,b) 2 / 59 mountain lion Distributional Semantics Models Aka, Vector Space Models, Word Embeddings 0 -0.23 1 0 -0.72 1 B -0.21 C B -00.2 C B C B C B -0.15 C B -0.71 C B C B C B -0.61 C B -0.13 C vmountain = B C, vlion = B C B . C B . C B . C B . C B C B C @ -0.02 A @ 0-0.1 A -0.12 -0.11 3 / 59 Distributional Semantics Models Aka, Vector Space Models, Word Embeddings 0 -0.23 1 0 -0.72 1 B -0.21 C B -00.2 C B C B C B -0.15 C B -0.71 C B C B C B -0.61 C B -0.13 C mountain vmountain = B C, vlion = B C B . C B . C B . C B . C B C B C lion @ -0.02 A @ 0-0.1 A -0.12 -0.11 4 / 59 Distributional Semantics Models Aka, Vector Space Models, Word Embeddings 0 -0.23 1 0 -0.72 1 B -0.21 C B -00.2 C B C B C B -0.15 C B -0.71 C B C B C B -0.61 C B -0.13 C mountain vmountain = B C, vlion = B C B . C B . C B .
    [Show full text]
  • Embeddings in Natural Language Processing
    Embeddings in Natural Language Processing Jose Camacho-Collados1 and Mohammad Taher Pilehvar2;3 1School of Computer Science and Informatics, Cardiff University, UK 2Tehran Institute for Advanced Studies (TeIAS), Tehran, Iran 3DTAL, University of Cambridge, UK [email protected], [email protected] Abstract Embeddings have been one of the most important topics of interest in Natural Language Process- ing (NLP) for the past decade. Representing knowledge through a low-dimensional vector which is easily integrable in modern machine learning models has played a central role in the develop- ment of the field. Embedding techniques initially focused on words but the attention soon started to shift to other forms. This tutorial will provide a high-level synthesis of the main embedding techniques in NLP, in the broad sense. We will start by conventional word embeddings (e.g., Word2Vec and GloVe) and then move to other types of embeddings, such as sense-specific and graph alternatives. We will finalize with an overview of the trending contextualized representa- tions (e.g., ELMo and BERT) and explain their potential and impact in NLP. 1 Description In this tutorial we will start by providing a historical overview on word-level vector space models, and word embeddings in particular. Word embeddings (e.g. Word2Vec (Mikolov et al., 2013), GloVe (Pen- nington et al., 2014) or FastText (Bojanowski et al., 2017)) have proven to be powerful keepers of prior knowledge to be integrated into downstream Natural Language Processing (NLP) applications. However, despite their flexibility and success in capturing semantic properties of words, the effec- tiveness of word embeddings are generally hampered by an important limitation, known as the meaning conflation deficiency: the inability to discriminate among different meanings of a word.
    [Show full text]
  • User Representation Learning for Social Networks: an Empirical Study
    applied sciences Article User Representation Learning for Social Networks: An Empirical Study Ibrahim Riza Hallac * , Betul Ay and Galip Aydin Department of Computer Engineering, Firat University, 23119 Elazig, Turkey; betulay@firat.edu.tr (B.A.); gaydin@firat.edu.tr (G.A.) * Correspondence: irhallac@firat.edu.tr Abstract: Gathering useful insights from social media data has gained great interest over the recent years. User representation can be a key task in mining publicly available user-generated rich content offered by the social media platforms. The way to automatically create meaningful observations about users of a social network is to obtain real-valued vectors for the users with user embedding representation learning models. In this study, we presented one of the most comprehensive studies in the literature in terms of learning high-quality social media user representations by leveraging state-of-the-art text representation approaches. We proposed a novel doc2vec-based representation method, which can encode both textual and non-textual information of a social media user into a low dimensional vector. In addition, various experiments were performed for investigating the performance of text representation techniques and concepts including word2vec, doc2vec, Glove, NumberBatch, FastText, BERT, ELMO, and TF-IDF. We also shared a new social media dataset comprising data from 500 manually selected Twitter users of five predefined groups. The dataset contains different activity data such as comment, retweet, like, location, as well as the actual tweets composed by the users. Keywords: distributed representation; user representation; user embedding; social networks Citation: Hallac, I.R.; Ay, B.; Aydin, G. User Representation Learning for Social Networks: An Empirical Study.
    [Show full text]
  • Learning Semantic Similarity in a Continuous Space
    Learning semantic similarity in a continuous space Michel Deudon Ecole Polytechnique Palaiseau, France [email protected] Abstract We address the problem of learning semantic representation of questions to measure similarity between pairs as a continuous distance metric. Our work naturally extends Word Mover’s Distance (WMD) [1] by representing text documents as normal distributions instead of bags of embedded words. Our learned metric measures the dissimilarity between two questions as the minimum amount of distance the intent (hidden representation) of one question needs to "travel" to match the intent of another question. We first learn to repeat, reformulate questions to infer intents as normal distributions with a deep generative model [2] (variational auto encoder). Semantic similarity between pairs is then learned discriminatively as an optimal transport distance metric (Wasserstein 2) with our novel variational siamese framework. Among known models that can read sentences individually, our proposed framework achieves competitive results on Quora duplicate questions dataset. Our work sheds light on how deep generative models can approximate distributions (semantic representations) to effectively measure semantic similarity with meaningful distance metrics from Information Theory. 1 Introduction Semantics is the study of meaning in language, used for understanding human expressions. It deals with the relationship between signifiers and what they stand for, their denotation. Measuring semantic similarity between pairs of sentences is an important problem in Natural Language Processing (NLP), for conversation systems (chatbots, FAQ), knowledge deduplication [3] or image captioning evaluation metrics [4] for example. In this paper, we consider the problem of determining the degree of similarity between pairs of sentences.
    [Show full text]
  • Evaluating Neural Word Representations in Tensor-Based
    Evaluating Neural Word Representations in Tensor-Based Compositional Settings Dmitrijs Milajevs1 Dimitri Kartsaklis2 Mehrnoosh Sadrzadeh1 Matthew Purver1 1Queen Mary University of London 2 School of Electronic Engineering University of Oxford and Computer Science Department of Computer Science Mile End Road, London, UK Parks Road, Oxford, UK [email protected] fd.milajevs,m.sadrzadeh,[email protected] Abstract ral word embeddings in compositional tasks, and meaningfully compare them with the traditional We provide a comparative study be- distributional approach based on co-occurrence tween neural word representations and counts. We are especially interested in investi- traditional vector spaces based on co- gating the performance of neural word vectors in occurrence counts, in a number of com- compositional models involving general mathe- positional tasks. We use three differ- matical composition operators, rather than in the ent semantic spaces and implement seven more task- or domain-specific deep-learning com- tensor-based compositional models, which positional settings they have generally been used we then test (together with simpler ad- with so far (for example, by Socher et al. (2012), ditive and multiplicative approaches) in Kalchbrenner and Blunsom (2013) and many oth- tasks involving verb disambiguation and ers). sentence similarity. To check their scala- bility, we additionally evaluate the spaces In particular, this is the first large-scale study using simple compositional methods on to date that applies neural word representations in larger-scale tasks with less constrained tensor-based compositional distributional models language: paraphrase detection and di- of meaning similar to those formalized by Coecke alogue act tagging.
    [Show full text]