DOCSLIB.ORG

Semantic Parsing on Freebase from Question-Answer Pairs

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Semantic Parsing on Freebase from Question-Answer Pairs Semantic Parsing on Freebase from Question-Answer Pairs Jonathan Berant Andrew Chou Roy Frostig Percy Liang Computer Science Department, Stanford University fjoberant,[email protected] frf,[email protected] Abstract Occidental College, Columbia University In this paper, we train a semantic parser that Execute on Database scales up to Freebase. Instead of relying on annotated logical forms, which is especially Type.University u Education:BarackObama expensive to obtain at large scale, we learn Type.University from question-answer pairs. The main chal- bridging lenge in this setting is narrowing down the Education huge number of possible logical predicates for alignment a given question. We tackle this problem in BarackObama two ways: First, we build a coarse mapping alignment from phrases to predicates using a knowledge Which college did Obama go to ? base and a large text corpus. Second, we use a bridging operation to generate additional Figure 1: Our task is to map questions to answers via la- predicates based on neighboring predicates. tent logical forms. To narrow down the space of logical On the dataset of Cai and Yates (2013), despite predicates, we use a (i) coarse alignment based on Free- not having annotated logical forms, our sys- base and a text corpus and (ii) a bridging operation that tem outperforms their state-of-the-art parser. generates predicates compatible with neighboring predi- Additionally, we collected a more realistic and cates. challenging dataset of question-answer pairs and improves over a natural baseline. predicates (Cai and Yates, 2013). The goal of this 1 Introduction paper is to do both: learn a semantic parser with- We focus on the problem of semantic parsing nat- out annotated logical forms that scales to the large ural language utterances into logical forms that can number of predicates on Freebase. be executed to produce denotations. Traditional se- At the lexical level, a major challenge in semantic mantic parsers (Zelle and Mooney, 1996; Zettle- parsing is mapping natural language phrases (e.g., moyer and Collins, 2005; Wong and Mooney, 2007; “attend”) to logical predicates (e.g., Education). Kwiatkowski et al., 2010) have two limitations: (i) While limited-domain semantic parsers are able they require annotated logical forms as supervision, to learn the lexicon from per-example supervision and (ii) they operate in limited domains with a small (Kwiatkowski et al., 2011; Liang et al., 2011), at number of logical predicates. Recent developments large scale they have inadequate coverage (Cai and aim to lift these limitations, either by reducing the Yates, 2013). Previous work on semantic parsing on amount of supervision (Clarke et al., 2010; Liang et Freebase uses a combination of manual rules (Yahya al., 2011; Goldwasser et al., 2011; Artzi and Zettle- et al., 2012; Unger et al., 2012), distant supervision moyer, 2011) or by increasing the number of logical (Krishnamurthy and Mitchell, 2012), and schema matching (Cai and Yates, 2013). We use a large 2.1 Knowledge base amount of web text and a knowledge base to build a Let E denote a set of entities (e.g., BarackObama), coarse alignment between phrases and predicates— and let P denote a set of properties (e.g., an approach similar in spirit to Cai and Yates (2013). PlaceOfBirth). A knowledge base K is a set However, this alignment only allows us to gen- of assertions (e1; p; e2) 2 E × P × E (e.g., erate a subset of the desired predicates. Aligning (BarackObama; PlaceOfBirth; Honolulu)). light verbs (e.g., “go”) and prepositions is not very We use the Freebase knowledge base (Google, informative due to polysemy, and rare predicates 2013), which has 41M non-numeric entities, 19K (e.g., “cover price”) are difficult to cover even given properties, and 596M assertions.1 a large corpus. To improve coverage, we propose a new bridging operation that generates predicates 2.2 Logical forms based on adjacent predicates rather than on words. At the compositional level, a semantic parser must To query the knowledge base, we use a logical lan- combine the predicates into a coherent logical form. guage called Lambda Dependency-Based Compo- Previous work based on CCG requires manually sitional Semantics (λ-DCS)—see Liang (2013) for specifying combination rules (Krishnamurthy and details. For the purposes of this paper, we use a re- Mitchell, 2012) or inducing the rules from anno- stricted subset called simple λ-DCS, which we will tated logical forms (Kwiatkowski et al., 2010; Cai define below for the sake of completeness. and Yates, 2013). We instead define a few simple The chief motivation of λ-DCS is to produce composition rules which over-generate and then use logical forms that are simpler than lambda cal- model features to simulate soft rules and categories. culus forms. For example, λx.9a:p1(x; a) ^ In particular, we use POS tag features and features 9b:p2(a; b) ^ p3(b; e) is expressed compactly in on the denotations of the predicted logical forms. λ-DCS as p1:p2:p3:e. Like DCS (Liang et al., We experimented with two question answering 2011), λ-DCS makes existential quantification im- datasets on Freebase. First, on the dataset of Cai plicit, thereby reducing the number of variables. and Yates (2013), we showed that our system out- Variables are only used for anaphora and building performs their state-of-the-art system 62% to 59%, composite binary predicates; these do not appear in despite using no annotated logical forms. Second, simple λ-DCS. we collected a new realistic dataset of questions by Each logical form in simple λ-DCS is either a performing a breadth-first search using the Google unary (which denotes a subset of E) or a binary Suggest API; these questions are then answered by (which denotes a subset of E × E). The basic λ- Amazon Mechanical Turk workers. Although this DCS logical forms z and their denotations z K are J K dataset is much more challenging and noisy, we are defined recursively as follows: still able to achieve 31.4% accuracy, a 4.5% ab- • Unary base case: If e 2 E is an entity (e.g., solute improvement over a natural baseline. Both Seattle), then e is a unary logical form with datasets as well as the source code for SEMPRE, our z K = feg. J K semantic parser, are publicly released and can be • Binary base case: If p 2 P is a property (e.g., downloaded from http://nlp.stanford.edu/ PlaceOfBirth), then p is a binary logical form 2 software/sempre/. with p K = f(e1; e2):(e1; p; e2) 2 Kg. • Join:J IfKb is a binary and u is a unary, then b:u 2 Setup (e.g., PlaceOfBirth.Seattle) is a unary de- noting a join and project: b:u K = fe1 2 E : J K Problem Statement Our task is as follows: Given 9e2:(e1; e2) 2 b K ^ e2 2 u Kg. (i) a knowledge base K, and (ii) a training set of J K J K n 1In this paper, we condense Freebase names for readability question-answer pairs f(xi; yi)g , output a se- i=1 /people/person Person mantic parser that maps new questions x to answers ( becomes ). 2Binaries can be also built out of lambda abstractions (e.g., y via latent logical forms z and the knowledge base λx.Performance:Actor:x), but as these constructions are K. not central to this paper, we defer to (Liang, 2013). • Intersection: If u1 and u2 are both unaries, Type.Location u PeopleBornHere.BarackObama intersection then u1 u u2 (e.g., Profession.Scientist u PlaceOfBirth.Seattle) denotes set intersec- Type.Location was PeopleBornHere.BarackObama ? lexicon join tion: u1 u u2 K = u1 K \ u2 K. • Aggregation:J K If u isJ aK unary,J thenK count(u) where BarackObama PeopleBornHere lexicon lexicon denotes the cardinality: count(u) K = J K Obama born fj u Kjg. As aJ finalK example, “number of dramas star- ring Tom Cruise” in lambda calculus would Figure 2: An example of a derivation d of the utterance be represented as count(λx.Genre(x; Drama) ^ “Where was Obama born?” and its sub-derivations, each 9y:Performance(x; y) ^ Actor(y; TomCruise)); labeled with composition rule (in blue) and logical form d “was” “?” in λ-DCS, it is simply count(Genre:Drama u (in red). The derivation skips the words and . Performance:Actor:TomCruise). It is useful to think of the knowledge base K as ily over-generates. We instead rely on features and a directed graph in which entities are nodes and learning to guide us away from the bad derivations. properties are labels on the edges. Then simple λ- DCS unary logical forms are tree-like graph patterns Modeling Following Zettlemoyer and Collins which pick out a subset of the nodes. (2005) and Liang et al. (2011), we define a discriminative log-linear model over derivations 2.3 Framework d 2 D(x) given utterances x: pθ(d j x) = expfφ(x;d)>θg x P 0 > , where φ(x; d) is a feature Given an utterance , our semantic parser constructs d02D(x) expfφ(x;d ) θg a distribution over possible derivations D(x). Each vector extracted from the utterance and the deriva- b derivation d 2 D(x) is a tree specifying the appli- tion, and θ 2 R is the vector of parameters to cation of a set of combination rules that culminates be learned. As our training data consists only of in the logical form d:z at the root of the tree—see question-answer pairs (xi; yi), we maximize the log- Figure 2 for an example.
Load more

Recommended publications
  • Semantic Parsing for Single-Relation Question Answering
    Semantic Parsing for Single-Relation Question Answering Wen-tau Yih Xiaodong He Christopher Meek Microsoft Research Redmond, WA 98052, USA scottyih,xiaohe,meek @microsoft.com { } Abstract the same question. That is to say that the problem of mapping from a question to a particular relation We develop a semantic parsing framework and entity in the KB is non-trivial. based on semantic similarity for open do- In this paper, we propose a semantic parsing main question answering (QA). We focus framework tailored to single-relation questions. on single-relation questions and decom- At the core of our approach is a novel semantic pose each question into an entity men- similarity model using convolutional neural net- tion and a relation pattern. Using convo- works. Leveraging the question paraphrase data lutional neural network models, we mea- mined from the WikiAnswers corpus by Fader et sure the similarity of entity mentions with al. (2013), we train two semantic similarity mod- entities in the knowledge base (KB) and els: one links a mention from the question to an the similarity of relation patterns and re- entity in the KB and the other maps a relation pat- lations in the KB. We score relational tern to a relation. The answer to the question can triples in the KB using these measures thus be derived by finding the relation–entity triple and select the top scoring relational triple r(e1, e2) in the KB and returning the entity not to answer the question. When evaluated mentioned in the question. By using a general se- on an open-domain QA task, our method mantic similarity model to match patterns and re- achieves higher precision across different lations, as well as mentions and entities, our sys- recall points compared to the previous ap- tem outperforms the existing rule learning system, proach, and can improve F1 by 7 points.
    [Show full text]
  • Semantic Parsing with Semi-Supervised Sequential Autoencoders
    Semantic Parsing with Semi-Supervised Sequential Autoencoders Toma´sˇ Kociskˇ y´†‡ Gabor´ Melis† Edward Grefenstette† Chris Dyer† Wang Ling† Phil Blunsom†‡ Karl Moritz Hermann† †Google DeepMind ‡University of Oxford tkocisky,melisgl,etg,cdyer,lingwang,pblunsom,kmh @google.com { } Abstract This is common in situations where we encode nat- ural language into a logical form governed by some We present a novel semi-supervised approach grammar or database. for sequence transduction and apply it to se- mantic parsing. The unsupervised component While such an autoencoder could in principle is based on a generative model in which latent be constructed by stacking two sequence transduc- sentences generate the unpaired logical forms. ers, modelling the latent variable as a series of dis- We apply this method to a number of semantic crete symbols drawn from multinomial distributions parsing tasks focusing on domains with lim- ited access to labelled training data and ex- creates serious computational challenges, as it re- tend those datasets with synthetically gener- quires marginalising over the space of latent se- ated logical forms. quences Σx∗. To avoid this intractable marginalisa- tion, we introduce a novel differentiable alternative for draws from a softmax which can be used with 1 Introduction the reparametrisation trick of Kingma and Welling (2014). Rather than drawing a discrete symbol in Neural approaches, in particular attention-based Σx from a softmax, we draw a distribution over sym- sequence-to-sequence models, have shown great bols from a logistic-normal distribution at each time promise and obtained state-of-the-art performance step. These serve as continuous relaxations of dis- for sequence transduction tasks including machine crete samples, providing a differentiable estimator translation (Bahdanau et al., 2015), syntactic con- of the expected reconstruction log likelihood.
    [Show full text]
  • SQL Generation from Natural Language: a Sequence-To- Sequence Model Powered by the Transformers Architecture and Association Rules
    Journal of Computer Science Original Research Paper SQL Generation from Natural Language: A Sequence-to- Sequence Model Powered by the Transformers Architecture and Association Rules 1,2Youssef Mellah, 1Abdelkader Rhouati, 1El Hassane Ettifouri, 2Toumi Bouchentouf and 2Mohammed Ghaouth Belkasmi 1NovyLab Research, Novelis, Paris, France 2Mohammed First University Oujda, National School of Applied Sciences, LaRSA/SmartICT Lab, Oujda, Morocco Article history Abstract: Using Natural Language (NL) to interacting with relational Received: 12-03-2021 databases allows users from any background to easily query and analyze Revised: 21-04-2021 large amounts of data. This requires a system that understands user Accepted: 28-04-2021 questions and automatically converts them into structured query language such as SQL. The best performing Text-to-SQL systems use supervised Corresponding Author: Youssef Mellah learning (usually formulated as a classification problem) by approaching NovyLab Research, Novelis, this task as a sketch-based slot-filling problem, or by first converting Paris, France questions into an Intermediate Logical Form (ILF) then convert it to the And corresponding SQL query. However, non-supervised modeling that LARSA Laboratory, ENSAO, directly converts questions to SQL queries has proven more difficult. In Mohammed First University, this sense, we propose an approach to directly translate NL questions into Oujda, Morocco SQL statements. In this study, we present a Sequence-to-Sequence Email: [email protected] (Seq2Seq) parsing model for the NL to SQL task, powered by the Transformers Architecture exploring the two Language Models (LM): Text-To-Text Transfer Transformer (T5) and the Multilingual pre-trained Text-To-Text Transformer (mT5). Besides, we adopt the transformation- based learning algorithm to update the aggregation predictions based on association rules.
    [Show full text]
  • Semantic Parsing with Neural Hybrid Trees
    Proceedings of the Thirty-First AAAI Conference on Artificial Intelligence (AAAI-17) Semantic Parsing with Neural Hybrid Trees Raymond Hendy Susanto, Wei Lu Singapore University of Technology and Design 8 Somapah Road, Singapore 487372 {raymond susanto, luwei}@sutd.edu.sg Abstract QUERY : answer(STATE) We propose a neural graphical model for parsing natural lan- STATE : exclude(STATE, STATE) guage sentences into their logical representations. The graph- ical model is based on hybrid tree structures that jointly rep- : state( ) : next to( ) resent both sentences and semantics. Learning and decoding STATE all STATE STATE are done using efficient dynamic programming algorithms. The model is trained under a discriminative setting, which STATE : stateid(STATENAME) allows us to incorporate a rich set of features. Hybrid tree structures have shown to achieve state-of-the-art results on STATENAME :(tx ) standard semantic parsing datasets. In this work, we propose a novel model that incorporates a rich, nonlinear featurization Which states do not border Texas ? by a feedforward neural network. The error signals are com- puted with respect to the conditional random fields (CRFs) objective using an inside-outside algorithm, which are then Figure 1: An example tree-structured semantic representa- backpropagated to the neural network. We demonstrate that tion and its corresponding natural language sentence. by combining the strengths of the exact global inference in the hybrid tree models and the power of neural networks to extract high level features, our model is able to achieve new explicit labelings between words and meaning representa- state-of-the-art results on standard benchmark datasets across tion tokens are not available in the training data.
    [Show full text]
  • Learning Logical Representations from Natural Languages with Weak Supervision and Back-Translation
    Learning Logical Representations from Natural Languages with Weak Supervision and Back-Translation Kaylin Hagopiany∗ Qing Wangy Georgia Institute of Technology IBM Research Atlanta, GA, USA Yorktown Heights, NY, USA [email protected] [email protected] Tengfei Ma Yupeng Gao Lingfei Wu IBM Research AI IBM Research AI IBM Research AI Yorktown Heights, NY, USA Yorktown Heights, NY, USA Yorktown Heights, NY, USA [email protected] [email protected] [email protected] Abstract Semantic parsing with strong supervision requires a large parallel corpus for train- ing, whose production may be impractical due to the excessive human labour needed for annotation. Recently, weakly supervised approaches based on reinforce- ment learning have been studied, wherein only denotations, but not the logical forms, are available. However, reinforcement learning approaches based on, e.g., policy gradient training, suffer reward sparsity and progress slowly. In this work, we explore enriching the policy gradient framework with a back-translation mecha- nism that boosts the reward signal. In this vein, we jointly optimize the expected reward of obtaining the correct denotation and the likelihood of reconstructing the natural language utterance. Initial experimental results suggest the proposed method slightly outperforms a state-of-the-art semantic parser trained with policy gradient. Introduction Semantic parsing is the task of converting natural language utterances to logical representations that a machine can understand and execute as problem solving steps. There has recently been a surge of interest in developing machine learning methods for semantic parsing, owing to the creation of large- scale text corpora. There has also been a proliferation of logical representations used in this setting, including logical forms [28], domain specific languages [26], python code [24], SQL queries [31], and bash commands [20].
    [Show full text]
  • A Sequence to Sequence Architecture for Task-Oriented Semantic Parsing
    Don’t Parse, Generate! A Sequence to Sequence Architecture for Task-Oriented Semantic Parsing Subendhu Rongali∗ Luca Soldaini University of Massachusetts Amherst Amazon Alexa Search Amherst, MA, USA Manhattan Beach, CA, USA [email protected] [email protected] Emilio Monti Wael Hamza Amazon Alexa Amazon Alexa AI Cambridge, UK New York, NY, USA [email protected] [email protected] ABSTRACT ArtistName Virtual assistants such as Amazon Alexa, Apple Siri, and Google Play the song don't stop believin by Journey Assistant often rely on a semantic parsing component to understand PlaySongIntent SongName which action(s) to execute for an utterance spoken by its users. Traditionally, rule-based or statistical slot-filling systems have been Intent: PlaySongIntent used to parse “simple” queries; that is, queries that contain a single Slots: SongName(don't stop believin), action and can be decomposed into a set of non-overlapping entities. ArtistName(Journey) More recently, shift-reduce parsers have been proposed to process more complex utterances. These methods, while powerful, impose specific limitations on the type of queries that can be parsed; namely, Figure 1: Semantic parsing of a “simple” query. Simple they require a query to be representable as a parse tree. queries define single action (intent) and can be decomposed In this work, we propose a unified architecture based on Se- into a set of non-overlapping entities (slots). quence to Sequence models and Pointer Generator Network to handle both simple and complex queries. Unlike other works, our 1 INTRODUCTION approach does not impose any restriction on the semantic parse Adoption of intelligent voice assistants such as Amazon Alexa, schema.
    [Show full text]
  • Neural Semantic Parsing Graham Neubig
    CS11-747 Neural Networks for NLP Neural Semantic Parsing Graham Neubig Site https://phontron.com/class/nn4nlp2018/ Tree Structures of Syntax • Dependency: focus on relations between words ROOT I saw a girl with a telescope • Phrase structure: focus on the structure of the sentence S VP PP NP NP PRP VBD DT NN IN DT NN I saw a girl with a telescope Representations of Semantics • Syntax only gives us the sentence structure • We would like to know what the sentence really means • Specifically, in an grounded and operationalizable way, so a machine can • Answer questions • Follow commands • etc. Meaning Representations • Special-purpose representations: designed for a specific task • General-purpose representations: designed to be useful for just about anything • Shallow representations: designed to only capture part of the meaning (for expediency) Parsing to Special-purpose Meaning Representations Example Special-purpose Representations • A database query language for sentence understanding • A robot command and control language • Source code in a language such as Python (?) Example Query Tasks • Geoquery: Parsing to Prolog queries over small database (Zelle and Mooney 1996) • Free917: Parsing to Freebase query language (Cai and Yates 2013) • Many others: WebQuestions, WikiTables, etc. Example Command and Control Tasks • Robocup: Robot command and control (Wong and Mooney 2006) • If this then that: Commands to smartphone interfaces (Quirk et al. 2015) Example Code Generation Tasks • Hearthstone cards (Ling et al. 2015) • Django commands (Oda
    [Show full text]
  • Practical Semantic Parsing for Spoken Language Understanding
    Practical Semantic Parsing for Spoken Language Understanding Marco Damonte∗ Rahul Goel Tagyoung Chung University of Edinburgh Amazon Alexa AI [email protected] fgoerahul,[email protected] Abstract the intent’s slots. This architecture is popular due to its simplicity and robustness. On the other hand, Executable semantic parsing is the task of con- verting natural language utterances into logi- Q&A, which need systems to produce more com- cal forms that can be directly used as queries plex structures such as trees and graphs, requires a to get a response. We build a transfer learn- more comprehensive understanding of human lan- ing framework for executable semantic pars- guage. ing. We show that the framework is effec- One possible system that can handle such a task tive for Question Answering (Q&A) as well as is an executable semantic parser (Liang, 2013; for Spoken Language Understanding (SLU). Kate et al., 2005). Given a user utterance, an exe- We further investigate the case where a parser on a new domain can be learned by exploit- cutable semantic parser can generate tree or graph ing data on other domains, either via multi- structures that represent logical forms that can be task learning between the target domain and used to query a knowledge base or database. In an auxiliary domain or via pre-training on the this work, we propose executable semantic pars- auxiliary domain and fine-tuning on the target ing as a common framework for both uses cases domain. With either flavor of transfer learn- by framing SLU as executable semantic parsing ing, we are able to improve performance on that unifies the two use cases.
    [Show full text]
  • Arxiv:2009.14116V1 [Cs.CL] 29 Sep 2020
    A Survey on Semantic Parsing from the Perspective of Compositionality Pawan Kumar Srikanta Bedathur Indian Institute of Technology Delhi, Indian Institute of Technology Delhi, New Delhi, Hauzkhas-110016 New Delhi, Hauzkhas-110016 [email protected] [email protected] Abstract compositionality: evident by the choice of the for- mal language and the construction mechanism and Different from previous surveys in seman- lexical variation: grounding the words/phrases to tic parsing (Kamath and Das, 2018) and appropriate KB entity/relation. knowledge base question answering(KBQA) (Chakraborty et al., 2019; Zhu et al., 2019; Language Compositionality: According to Hoffner¨ et al., 2017) we try to takes a dif- Frege’s principle of compositionality: “the ferent perspective on the study of semantic meaning of a whole is a function of the meaning parsing. Specifically, we will focus on (a) meaning composition from syntactical struc- of the parts. The study of formal semantics ture (Partee, 1975), and (b) the ability of for natural language to appropriately represent semantic parsers to handle lexical variation sentence meaning has a long history (Szab, 2017). given the context of a knowledge base (KB). Most semantic formalism follow the tradition of In the following section after an introduction Montague’s grammar (Partee, 1975) i.e. there is of the field of semantic parsing and its uses a one-to-one correspondence between syntax and in KBQA, we will describe meaning represen- semantics e.g. CCG (Zettlemoyer and Collins, tation using grammar formalism CCG (Steed- 2005). We will not be delving into the concept of man, 1996).
    [Show full text]
  • Learning Contextual Representations for Semantic Parsing with Generation-Augmented Pre-Training
    Learning Contextual Representations for Semantic Parsing with Generation-Augmented Pre-Training Peng Shi1*, Patrick Ng2, Zhiguo Wang2, Henghui Zhu2, Alexander Hanbo Li2, Jun Wang2, Cicero Nogueira dos Santos2, Bing Xiang2 1 University of Waterloo, 2 AWS AI Labs [email protected], fpatricng,zhiguow,henghui,hanboli,juwanga,cicnog,[email protected] Abstract Pain Point 1: Fail to match and detect the column mentions. Utterance: Which professionals live in a city containing the Most recently, there has been significant interest in learn- substring ’West’? List his or her role, street, city and state. ing contextual representations for various NLP tasks, by Prediction: SELECT role code, street, state leveraging large scale text corpora to train large neural lan- FROM Professionals WHERE city LIKE ’%West%’ guage models with self-supervised learning objectives, such Error: Missing column city in SELECT clause. as Masked Language Model (MLM). However, based on a pi- lot study, we observe three issues of existing general-purpose Pain Point 2: Fail to infer columns based on cell values. language models when they are applied to text-to-SQL se- Utterance: Give the average life expectancy for countries in mantic parsers: fail to detect column mentions in the utter- Africa which are republics? ances, fail to infer column mentions from cell values, and fail Prediction: SELECT Avg(LifeExpectancy) FROM to compose complex SQL queries. To mitigate these issues, country WHERE Continent = ’Africa’ we present a model pre-training framework, Generation- Error: Missing GovernmentForm = ’Republic’. Augmented Pre-training (GAP), that jointly learns repre- Pain Point 3 sentations of natural language utterances and table schemas : Fail to compose complex target SQL.
    [Show full text]
  • On the Use of Parsing for Named Entity Recognition
    applied sciences Review On the Use of Parsing for Named Entity Recognition Miguel A. Alonso * , Carlos Gómez-Rodríguez and Jesús Vilares Grupo LyS, Departamento de Ciencias da Computación e Tecnoloxías da Información, Universidade da Coruña and CITIC, 15071 A Coruña, Spain; [email protected] (C.G.-R.); [email protected] (J.V.) * Correspondence: [email protected] Abstract: Parsing is a core natural language processing technique that can be used to obtain the structure underlying sentences in human languages. Named entity recognition (NER) is the task of identifying the entities that appear in a text. NER is a challenging natural language processing task that is essential to extract knowledge from texts in multiple domains, ranging from financial to medical. It is intuitive that the structure of a text can be helpful to determine whether or not a certain portion of it is an entity and if so, to establish its concrete limits. However, parsing has been a relatively little-used technique in NER systems, since most of them have chosen to consider shallow approaches to deal with text. In this work, we study the characteristics of NER, a task that is far from being solved despite its long history; we analyze the latest advances in parsing that make its use advisable in NER settings; we review the different approaches to NER that make use of syntactic information; and we propose a new way of using parsing in NER based on casting parsing itself as a sequence labeling task. Keywords: natural language processing; named entity recognition; parsing; sequence labeling Citation: Alonso, M.A; Gómez- 1.
    [Show full text]
  • Neural Semantic Parsing for Syntax-Aware Code Generation
    Neural Semantic Parsing for Syntax-Aware Code Generation Masterarbeit zur Erlangung des Master-Grades Master of Science im Studiengang Tech- nische Informatik an der Fakultät für Informations-, Medien- und Elektrotechnik der Technischen Hochschule Köln vorgelegt von: Artur Baranowski Matrikel-Nr.: 11130536 Adresse: Heidestraße 144 51147 Köln Erstprüfer: Prof. Dr. Hubert Randerath Zweitprüfer: Prof. Dr. Nico Hochgeschwender Köln, 23. September 2020 Abstract I Abstract The task of mapping natural language expressions to logical forms is referred to as semantic parsing. The syntax of logical forms that are based on programming or query languages, such as Python or SQL, is defined by a formal grammar. In this thesis, we present an efficient neural semantic parser that exploits the underlying grammar of logical forms to enforce well-formed expressions. We use an encoder-decoder model for sequence prediction. Syntactically valid programs are guaranteed by means of a bottom- up shift-reduce parser, that keeps track of the set of viable tokens at each decoding step. We show that the proposed model outperforms the standard encoder-decoder model across datasets and is competitive with comparable grammar-guided semantic parsing approaches. Keywords: semantic parsing, machine learning, natural language processing, code generation, formal grammars, parsing theory Statutory Declaration II Statutory Declaration I herewith declare that I have composed the present thesis myself and without use of any other than the cited sources and aids. Sentences or parts of sentences quoted literally are marked as such; other references with regard to the statement and scope are indicated by full details of the publications concerned. The thesis in the same or similar form has not been submitted to any examination body and has not been published.
    [Show full text]