Natural language understanding is a sub-field of natural language processing, which builds automated systems to understand natural language. It is such an ambitious task that it sometimes is referred to as an AI-complete problem, implying that its difficulty is equivalent to solving the central artificial intelligence problem -- making computers as intelligent as people. Despite its complexity, natural language understanding continues to be a fundamental problem in natural language processing in terms of its theoretical and empirical importance. In recent years, startling progress has been made at different levels of natural language processing tasks, which provides great opportunity for deeper natural language understanding. In this thesis, we focus on the task of semantic parsing, which maps a natural language sentence into a complete, formal meaning representation in a meaning representation language. We present two novel state-of-the-art learned syntax-based semantic parsers using statistical syntactic parsing techniques, motivated by the following two reasons. First, the syntax-based semantic parsing is theoretically well-founded in computational semantics. Second, adopting a syntax-based approach allows us to directly leverage the enormous progress made in statistical syntactic parsing. The first semantic parser, Scissor, adopts an integrated syntactic-semantic parsing approach, in which a statistical syntactic parser is augmented with semantic parameters to produce a semantically-augmented parse tree (SAPT). This integrated approach allows both syntactic and semantic information to be available during parsing time to obtain an accurate combined syntactic-semantic analysis. The performance of Scissor is further improved by using discriminative reranking for incorporating non-local features. The second semantic parser, SynSem, exploits an existing syntactic parser to produce disambiguated parse trees that drive the compositional semantic interpretation. This pipeline approach allows semantic parsing to conveniently leverage the most recent progress in statistical syntactic parsing. We report experimental results on two real applications: an interpreter for coaching instructions in robotic soccer and a natural-language database interface, showing that the improvement of Scissor and SynSem over other systems is mainly on long sentences, where the knowledge of syntax given in the form of annotated SAPTs or syntactic parses from an existing parser helps semantic composition. SynSem also significantly improves results with limited training data, and is shown to be robust to syntactic errors. / text

Querying a database to retrieve an answer, telling a robot to perform an action, or teaching a computer to play a game are tasks requiring communication with machines in a language interpretable by them. Semantic parsing is the task of converting human language to a machine interpretable language. While human languages are sequential in nature with latent structures, machine interpretable languages are formal with explicit structures. The computational linguistics community have created several treebanks to understand the formal syntactic structures of human languages. In this thesis, we use these to obtain formal meaning representations of languages, and learn computational models to convert these meaning representations to the target machine representation. Our goal is to evaluate if existing treebank syntactic representations are useful for semantic parsing. Existing semantic parsing methods mainly learn domain-specific grammars which can parse human languages to machine representation directly. We deviate from this trend and make use of general-purpose syntactic grammar to help in semantic parsing. We use two syntactic representations: Combinatory Categorial Grammar (CCG) and dependency syntax. CCG has a well established theory on deriving meaning representations from its syntactic derivations. But there are no CCG treebanks for many languages since these are difficult to annotate. In contrast, dependencies are easy to annotate and have many treebanks. However, dependencies do not have a well established theory for deriving meaning representations. In this thesis, we propose novel theories for deriving meaning representations from dependencies. Our evaluation task is question answering on a knowledge base. Given a question, our goal is to answer it on the knowledge base by converting the question to an executable query. We use Freebase, the knowledge source behind Google’s search engine, as our knowledge base. Freebase contains millions of real world facts represented in a graphical format. Inspired from the Freebase structure, we formulate semantic parsing as a graph matching problem, i.e., given a natural language sentence, we convert it into a graph structure from the meaning representation obtained from syntax, and find the subgraph of Freebase that best matches the natural language graph. Our experiments on Free917, WebQuestions and GraphQuestions semantic parsing datasets conclude that general-purpose syntax is more useful for semantic parsing than induced task-specific syntax and syntax-agnostic representations.

Notre but dans cette thèse est de construire un système qui réponde à une question en langue naturelle (NL) en représentant sa sémantique comme une forme logique (LF) et ensuite en calculant une réponse en exécutant cette LF sur une base de connaissances. La partie centrale d'un tel système est l'analyseur sémantique qui transforme les questions en formes logiques. Notre objectif est de construire des analyseurs sémantiques performants en apprenant à partir de paires (NL, LF). Nous proposons de combiner des réseaux neuronaux récurrents (RNN) avec des connaissances préalables symboliques exprimées à travers des grammaires hors-contexte (CFGs) et des automates. En intégrant des CFGs contrôlant la validité des LFs dans les processus d'apprentissage et d'inférence des RNNs, nous garantissons que les formes logiques générées sont bien formées; en intégrant, par le biais d'automates pondérés, des connaissances préalables sur la présence de certaines entités dans la LF, nous améliorons encore la performance de nos modèles. Expérimentalement, nous montrons que notre approche permet d'obtenir de meilleures performances que les analyseurs sémantiques qui n'utilisent pas de réseaux neuronaux, ainsi que les analyseurs à base de RNNs qui ne sont pas informés par de telles connaissances préalables / Our goal in this thesis is to build a system that answers a natural language question (NL) by representing its semantics as a logical form (LF) and then computing the answer by executing the LF over a knowledge base. The core part of such a system is the semantic parser that maps questions to logical forms. Our focus is how to build high-performance semantic parsers by learning from (NL, LF) pairs. We propose to combine recurrent neural networks (RNNs) with symbolic prior knowledge expressed through context-free grammars (CFGs) and automata. By integrating CFGs over LFs into the RNN training and inference processes, we guarantee that the generated logical forms are well-formed; by integrating, through weighted automata, prior knowledge over the presence of certain entities in the LF, we further enhance the performance of our models. Experimentally, we show that our approach achieves better performance than previous semantic parsers not using neural networks as well as RNNs not informed by such prior knowledge

This thesis presents novel tasks and deep learning methods for machine reading comprehension and question answering with the goal of achieving natural language understanding. First, we consider a semantic parsing task where the model understands sentences and translates them into a logical form or instructions. We present a novel semi-supervised sequential autoencoder that considers language as a discrete sequential latent variable and semantic parses as the observations. This model allows us to leverage synthetically generated unpaired logical forms, and thereby alleviate the lack of supervised training data. We show the semi-supervised model outperforms a supervised model when trained with the additional generated data. Second, reading comprehension requires integrating information and reasoning about events, entities, and their relations across a full document. Question answering is conventionally used to assess reading comprehension ability, in both artificial agents and children learning to read. We propose a new, challenging, supervised reading comprehension task. We gather a large-scale dataset of news stories from the CNN and Daily Mail websites with Cloze-style questions created from the highlights. This dataset allows for the first time training deep learning models for reading comprehension. We also introduce novel attention-based models for this task and present qualitative analysis of the attention mechanism. Finally, following the recent advances in reading comprehension in both models and task design, we further propose a new task for understanding complex narratives, NarrativeQA, consisting of full texts of books and movie scripts. We collect human written questions and answers based on high-level plot summaries. This task is designed to encourage development of models for language understanding; it is designed so that successfully answering their questions requires understanding the underlying narrative rather than relying on shallow pattern matching or salience. We show that although humans solve the tasks easily, standard reading comprehension models struggle on the tasks presented here.

Communicating with natural language interfaces is a long-standing, ultimate goal for artificial intelligence (AI) agents to pursue, eventually. One core issue toward this goal is "grounded" language learning, a process of learning the semantics of natural language with respect to relevant perceptual inputs. In order to ground the meanings of language in a real world situation, computational systems are trained with data in the form of natural language sentences paired with relevant but ambiguous perceptual contexts. With such ambiguous supervision, it is required to resolve the ambiguity between a natural language (NL) sentence and a corresponding set of possible logical meaning representations (MR). In this thesis, we focus on devising effective models for simultaneously disambiguating such supervision and learning the underlying semantics of language to map NL sentences into proper logical MRs. We present probabilistic generative models for learning such correspondences along with a reranking model to improve the performance further. First, we present a probabilistic generative model that learns the mappings from NL sentences into logical forms where the true meaning of each NL sentence is one of a handful of candidate logical MRs. It simultaneously disambiguates the meaning of each sentence in the training data and learns to probabilistically map an NL sentence to its corresponding MR form depicted in a single tree structure. We perform evaluations on the RoboCup sportscasting corpus, proving that our model is more effective than those proposed by previous researchers. Next, we describe two PCFG induction models for grounded language learning that extend the previous grounded language learning model of Börschinger, Jones, and Johnson (2011). Börschinger et al.’s approach works well in situations of limited ambiguity, such as in the sportscasting task. However, it does not scale well to highly ambiguous situations when there are large sets of potential meaning possibilities for each sentence, such as in the navigation instruction following task first studied by Chen and Mooney (2011). The two models we present overcome such limitations by employing a learned semantic lexicon as a basic correspondence unit between NL and MR for PCFG rule generation. Finally, we present a method of adapting discriminative reranking to grounded language learning in order to improve the performance of our proposed generative models. Although such generative models are easy to implement and are intuitive, it is not always the case that generative models perform best, since they are maximizing the joint probability of data and model, rather than directly maximizing conditional probability. Because we do not have gold-standard references for training a secondary conditional reranker, we incorporate weak supervision of evaluations against the perceptual world during the process of improving model performance. All these approaches are evaluated on the two publicly available domains that have been actively used in many other grounded language learning studies. Our methods demonstrate consistently improved performance over those of previous studies in the domains with different languages; this proves that our methods are language-independent and can be generally applied to other grounded learning problems as well. Further possible applications of the presented approaches include summarized machine translation tasks and learning from real perception data assisted by computer vision and robotics. / text

The cross-situational word learning paradigm argues that word meanings can be approximated by word-object associations, computed from co-occurrence statistics between words and entities in the world. Lexicon acquisition involves simultaneously guessing (1) which objects are being talked about (the ”meaning”) and (2) which words relate to those objects. However, most modeling work focuses on acquiring meanings for isolated words, largely neglecting relationships between words or physical entities, which can play an important role in learning. Semantic parsing, on the other hand, aims to learn a mapping between entire utterances and compositional meaning representations where such relations are central. The focus is the mapping between meaning and words, while utterance meanings are treated as observed quantities. Here, we extend the joint inference problem of word learning to account for compositional meanings by incorporating a semantic parsing model for relating utterances to non-linguistic context. Integrating semantic parsing and word learning permits us to explore the impact of word-word and concept-concept relations. The result is a joint-inference problem inherited from the word learning setting where we must simultaneously learn utterance-level and individual word meanings, only now we also contend with the many possible relationships between concepts in the meaning and words in the sentence. To simplify design, we factorize the model into separate modules, one for each of the world, the meaning, and the words, and merge them into a single synchronous grammar for joint inference. There are three main contributions. First, we introduce a novel word learning model and accompanying semantic parser. Second, we produce a corpus which allows us to demonstrate the importance of structure in word learning. Finally, we also present a number of technical innovations required for implementing such a model.

No description available.

abstract: Turing test has been a benchmark scale for measuring the human level intelligence in computers since it was proposed by Alan Turing in 1950. However, for last 60 years, the applications such as ELIZA, PARRY, Cleverbot and Eugene Goostman, that claimed to pass the test. These applications are either based on tricks to fool humans on a textual chat based test or there has been a disagreement between AI communities on them passing the test. This has led to the school of thought that it might not be the ideal test for predicting the human level intelligence in machines. Consequently, the Winograd Schema Challenge has been suggested as an alternative to the Turing test. As opposed to deciding the intelligent behavior with the help of chat servers, like it was done in the Turing test, the Winograd Schema Challenge is a question answering test. It consists of sentence and question pairs such that the answer to the question depends on the resolution of a definite pronoun or adjective in the sentence. The answers are fairly intuitive for humans but they are difficult for machines because it requires some sort of background or commonsense knowledge about the sentence. In this thesis, I propose a novel technique to solve the Winograd Schema Challenge. The technique has three basic modules at its disposal, namely, a Semantic Parser that parses the English text (both sentences and questions) into a formal representation, an Automatic Background Knowledge Extractor that extracts the Background Knowledge pertaining to the given Winograd sentence, and an Answer Set Programming Reasoning Engine that reasons on the given Winograd sentence and the corresponding Background Knowledge. The applicability of the technique is illustrated by solving a subset of Winograd Schema Challenge pertaining to a certain type of Background Knowledge. The technique is evaluated on the subset and a notable accuracy is achieved. / Dissertation/Thesis / Masters thesis defense presentation slides / Masters Thesis Computer Science 2014

No description available.

Deep learning has been successfully applied to semantic graph parsing in recent years. However, to our best knowledge, all graph-based parsers depend on a strong assumption about the ordering of graph nodes. This work explores a permutation-invariant approach to sentence-to-graph semantic parsing. We present a versatile, cross-framework, and language-independent architecture for universal modeling of semantic structures. To empirically validate our method, we participated in the CoNLL 2020 shared task, Cross- Framework Meaning Representation Parsing (MRP 2020), which evaluated the competing systems on five different frameworks (AMR, DRG, EDS, PTG, and UCCA) across four languages. Our parsing system, called PERIN, was one of the winners of this shared task. Thus, we believe that permutation invariance is a promising new direction in the field of semantic parsing. 1