Showing papers on "Natural language understanding published in 2015"

Do Multi-Sense Embeddings Improve Natural Language Understanding?

Abstract: Learning a distinct representation for each sense of an ambiguous word could lead to more powerful and fine-grained models of vector-space representations. Yet while ‘multi-sense’ methods have been proposed and tested on artificial wordsimilarity tasks, we don’t know if they improve real natural language understanding tasks. In this paper we introduce a multisense embedding model based on Chinese Restaurant Processes that achieves state of the art performance on matching human word similarity judgments, and propose a pipelined architecture for incorporating multi-sense embeddings into language understanding. We then test the performance of our model on part-of-speech tagging, named entity recognition, sentiment analysis, semantic relation identification and semantic relatedness, controlling for embedding dimensionality. We find that multi-sense embeddings do improve performance on some tasks (part-of-speech tagging, semantic relation identification, semantic relatedness) but not on others (named entity recognition, various forms of sentiment analysis). We discuss how these differences may be caused by the different role of word sense information in each of the tasks. The results highlight the importance of testing embedding models in real applications.

Do Multi-Sense Embeddings Improve Natural Language Understanding?

Abstract: Learning a distinct representation for each sense of an ambiguous word could lead to more powerful and fine-grained models of vector-space representations. Yet while `multi-sense' methods have been proposed and tested on artificial word-similarity tasks, we don't know if they improve real natural language understanding tasks. In this paper we introduce a multi-sense embedding model based on Chinese Restaurant Processes that achieves state of the art performance on matching human word similarity judgments, and propose a pipelined architecture for incorporating multi-sense embeddings into language understanding. We then test the performance of our model on part-of-speech tagging, named entity recognition, sentiment analysis, semantic relation identification and semantic relatedness, controlling for embedding dimensionality. We find that multi-sense embeddings do improve performance on some tasks (part-of-speech tagging, semantic relation identification, semantic relatedness) but not on others (named entity recognition, various forms of sentiment analysis). We discuss how these differences may be caused by the different role of word sense information in each of the tasks. The results highlight the importance of testing embedding models in real applications.

Reasoning about Quantities in Natural Language

Abstract: Little work from the Natural Language Processing community has targeted the role of quantities in Natural Language Understanding. This paper takes some key steps towards facilitating reasoning about quantities expressed in natural language. We investigate two different tasks of numerical reasoning. First, we consider Quantity Entailment, a new task formulated to understand the role of quantities in general textual inference tasks. Second, we consider the problem of automatically understanding and solving elementary school math word problems. In order to address these quantitative reasoning problems we first develop a computational approach which we show to successfully recognize and normalize textual expressions of quantities. We then use these capabilities to further develop algorithms to assist reasoning in the context of the aforementioned tasks.

Voice and connection platform

Abstract: A system and method for providing a voice assistant including receiving, at a first device, a first audio input from a user requesting a first action; performing automatic speech recognition on the first audio input; obtaining a context of user; performing natural language understanding based on the speech recognition of the first audio input; and taking the first action based on the context of the user and the natural language understanding.

A survey of qualitative spatial representations

Abstract: Representation and reasoning with qualitative spatial relations is an important problem in artificial intelligence and has wide applications in the fields of geographic information system, computer vision, autonomous robot navigation, natural language understanding, spatial databases and so on. The reasons for this interest in using qualitative spatial relations include cognitive comprehensibility, efficiency and computational facility. This paper summarizes progress in qualitative spatial representation by describing key calculi representing different types of spatial relationships. The paper concludes with a discussion of current research and glimpse of future work.

Generating High Quality Proposition Banks for Multilingual Semantic Role Labeling

Abstract: Semantic role labeling (SRL) is crucial to natural language understanding as it identifies the predicate-argument structure in text with semantic labels. Unfortunately, resources required to construct SRL models are expensive to obtain and simply do not exist for most languages. In this paper, we present a two-stage method to enable the construction of SRL models for resourcepoor languages by exploiting monolingual SRL and multilingual parallel data. Experimental results show that our method outperforms existing methods. We use our method to generate Proposition Banks with high to reasonable quality for 7 languages in three language families and release these resources to the research community.

Natural Language Understanding with Distributed Representation

Abstract: This is a lecture note for the course DS-GA 3001 at the Center for Data Science , New York University in Fall, 2015. As the name of the course suggests, this lecture note introduces readers to a neural network based approach to natural language understanding/processing. In order to make it as self-contained as possible, I spend much time on describing basics of machine learning and neural networks, only after which how they are used for natural languages is introduced. On the language front, I almost solely focus on language modelling and machine translation, two of which I personally find most fascinating and most fundamental to natural language understanding.

Provably correct reactive control from natural language

Abstract: This paper presents an integrated system for generating, troubleshooting, and executing correct-by-construction controllers for autonomous robots using natural language input, allowing non-expert users to command robots to perform high-level tasks. This system unites the power of formal methods with the accessibility of natural language, providing controllers for implementable high-level task specifications, easy-to-understand feedback on those that cannot be achieved, and natural language explanation of the reason for the robot's actions during execution. The natural language system uses domain-general components that can easily be adapted to cover the vocabulary of new applications. Generation of a linear temporal logic specification from the user's natural language input uses a novel data structure that allows for subsequent mapping of logical propositions back to natural language, enabling natural language feedback about problems with the specification that are only identifiable in the logical form. We demonstrate the robustness of the natural language understanding system through a user study where participants interacted with a simulated robot in a search and rescue scenario. Automated analysis and user feedback on unimplementable specifications is demonstrated using an example involving a robot assistant in a hospital.

A Tableau Prover for Natural Logic and Language

Abstract: Modeling the entailment relation over sentences is one of the generic problems of natural language understanding. In order to account for this problem, we design a theorem prover for Natural Logic, a logic whose terms resemble natural language expressions. The prover is based on an analytic tableau method and employs syntactically and semantically motivated schematic rules. Pairing the prover with a preprocessor, which generates formulas of Natural Logic from linguistic expressions, results in a proof system for natural language. It is shown that the system obtains a comparable accuracy (81%) on the unseen SICK data while achieving the state-of-the-art precision (98%).

A Framework for Procedural Text Understanding

Abstract: In this paper we propose a framework for procedural text understanding. Procedural texts are relatively clear without modality nor dependence on viewpoints, etc. and have many potential applications in artificial intelligence. Thus they are suitable as the first target of natural language understanding. As our framework we extend parsing technologies to connect important concepts in a text. Our framework first tokenizes the input text, a sequence of sentences, then recognizes important concepts like named entity recognition, and finally connect them like a sentence parser but dealing all the concepts in the text at once. We tested our framework on cooking recipe texts annotated with a directed acyclic graph as their meaning. We present experimental results and evaluate our framework.

Recognizing Textual Entailment

Abstract: Since 2005, researchers have worked on a broad task called Recognizing Textual Entailment (RTE), which is designed to focus efforts on general textual inference capabilities, but without constraining participants to use a specific representation or reasoning approach. There have been promising developments in this sub-field of Natural Language Processing (NLP), with systems showing steady improvement, and investigations of a range of approaches to the problem. A number of researchers appear to have converged on some defining characteristics of the problem, and on characteristics of practical approaches to solving it. RTE solutions have been shown to be of practical use in other NLP applications, and other grand Natural Language Understanding (NLU) challenges, such as Learning by Reading [27] and Machine Reading [43] have emerged that will require similar problems to be solved. It is an exciting time to be working in this area. Textual Inference is a key capability for improving performance in a wide range of NLP tasks, particularly those which can benefit from integrating background knowledge. Performance of Question-Answering systems, which can be thought of as potentially the next generation of search engines, is limited, especially outside the class of factoid questions; and the task of extracting facts of interest (such as “People who have worked for Company X”) from a collection of plain text documents (such as newspaper articles) may require significant abstraction, synthesis, and application of world knowledge on the part of a human reader – and therefore

Natural language understanding method and travel question-answering system based on same

Abstract: The invention discloses a natural language understanding method and a travel question-answering system based on the same. The natural language understanding method is characterized in that questions are matched with questions in a syntax database, answers are extracted from a knowledge database according to functions and parameters corresponding to the questions, the syntax database contains most questions in the domain, questions out of the domain are provided through non-domain knowledge databases, the acquired answers are allowed to conform to the context when a user asks the questions by reading the related data of historical questions in a cache, and the accurate answers and related answers are provided. When the method is used for the travel question-answering system, more than 99% pf questions in the travel domain can be covered, the answers out of the travel domain can be extracted, the related answers can be provided, and tests show that accuracy can reach more than 95%.

Architecture for multi-domain natural language processing

Abstract: Features are disclosed for processing a user utterance with respect to multiple subject matters or domains, and for selecting a likely result from a particular domain with which to respond to the utterance or otherwise take action. A user utterance may be transcribed by an automatic speech recognition (“ASR”) module, and the results may be provided to a multi-domain natural language understanding (“NLU”) engine. The multi-domain NLU engine may process the transcription(s) in multiple individual domains rather than in a single domain. In some cases, the transcription(s) may be processed in multiple individual domains in parallel or substantially simultaneously. In addition, hints may be generated based on previous user interactions and other data. The ASR module, multi-domain NLU engine, and other components of a spoken language processing system may use the hints to more efficiently process input or more accurately generate output.

Understanding Student Language: An Unsupervised Dialogue Act Classification Approach

Abstract: Within the landscape of educational data, textual natural language is an increasingly vast source of learning-centered interactions In natural language dialogue, student contributions hold important information about knowledge and goals Automatically modeling the dialogue act of these student utterances is crucial for scaling natural language understanding of educational dialogues Automatic dialogue act modeling has long been addressed with supervised classification techniques that require substantial manual time and effort Recently, there is emerging interest in unsupervised dialogue act classification, which addresses the challenges related to manually labeling corpora This paper builds on the growing body of work in unsupervised dialogue act classification and reports on the novel application of an information retrieval technique, the Markov Random Field, for the task of unsupervised dialogue act classification Evaluation against manually labeled dialogue acts on a tutorial dialogue corpus in the domain of introductory computer science demonstrates that the proposed technique outperforms existing approaches to education-centered unsupervised dialogue act classification Unsupervised dialogue act classification techniques have broad application in educational data mining in areas such as collaborative learning, online message boards, classroom discourse, and intelligent tutoring systems

Segment-Phrase Table for Semantic Segmentation, Visual Entailment and Paraphrasing

Abstract: We introduce Segment-Phrase Table (SPT), a large collection of bijective associations between textual phrases and their corresponding segmentations. Leveraging recent progress in object recognition and natural language semantics, we show how we can successfully build a high-quality segment-phrase table using minimal human supervision. More importantly, we demonstrate the unique value unleashed by this rich bimodal resource, for both vision as well as natural language understanding. First, we show that fine-grained textual labels facilitate contextual reasoning that helps in satisfying semantic constraints across image segments. This feature enables us to achieve state-of-the-art segmentation results on benchmark datasets. Next, we show that the association of high-quality segmentations to textual phrases aids in richer semantic understanding and reasoning of these textual phrases. Leveraging this feature, we motivate the problem of visual entailment and visual paraphrasing, and demonstrate its utility on a large dataset.

Visual Commonsense for Scene Understanding Using Perception, Semantic Parsing and Reasoning.

Abstract: In this paper we explore the use of visual commonsense knowledge and other kinds of knowledge (such as domain knowledge, background knowledge, linguistic knowledge) for scene understanding. In particular, we combine visual processing with techniques from natural language understanding (especially semantic parsing), common-sense reasoning and knowledge representation and reasoning to improve visual perception to reason about finer aspects of activities.

Scaling Semantic Frame Annotation

Abstract: Large-scale data resources needed for progress toward natural language understanding are not yet widely available and typically require considerable expense and expertise to create. This paper addresses the problem of developing scalable approaches to annotating semantic frames and explores the viability of crowdsourcing for the task of frame disambiguation. We present a novel supervised crowdsourcing paradigm that incorporates insights from human computation research designed to accommodate the relative complexity of the task, such as exemplars and real-time feedback. We show that non-experts can be trained to perform accurate frame disambiguation, and can even identify errors in gold data used as the training exemplars. Results demonstrate the efficacy of this paradigm for semantic annotation requiring an intermediate level of expertise. 1 The semantic bottleneck Behind every great success in speech and language lies a great corpus—or at least a very large one. Advances in speech recognition, machine translation and syntactic parsing can be traced to the availability of large-scale annotated resources (Wall Street Journal, Europarl and Penn Treebank, respectively) providing crucial supervised input to statistically learned models. Semantically annotated resources have been comparatively harder to come by: representing meaning poses myriad philosophical, theoretical and practical challenges, particularly for general purpose resources that can be applied to diverse domains. If these challenges can be addressed, however, semantic resources hold significant potential for fueling progress beyond shallow syntax and toward deeper language understanding. This paper explores the feasibility of developing scalable methodologies for semantic annotation, inspired by three strands of work. First, frame semantics, and its instantiation in the Berkeley FrameNet project (Fillmore and Baker, 2010), offers a principled approach to representing meaning. FrameNet is a lexicographic resource that captures syntactic and semantic generalizations that go beyond surface form and part of speech, famously including the relationships among words like buy, sell, purchase and price. These rich structural relations provide an attractive foundation for work in deeper natural language understanding and inference, as attested by the breadth of applications at the Workshop in Honor of Chuck Fillmore at ACL 2014 (Petruck and de Melo, 2014). But FrameNet was not designed to support scalable language technologies; indeed, it is perhaps a paradigm example of a hand-curated knowledge resource, one that has required significant expertise, training, time and expense to create and that remains under development. Second, the task of automatic semantic role labeling (ASRL) (Gildea and Jurafsky, 2002) serves as an applied counterpart to the ideas of frame semantics. Recent progress has demonstrated the viability of training automated models using frameannotated data (Das et al., 2013; Das et al., 2010; Johansson and Nugues, 2006). Results based on FrameNet data have been limited by its incomplete

Multi-purpose natural language understanding linked to sensorimotor experience in humanoid robots

Abstract: Humans have an amazing ability to bootstrap new knowledge. The concept of structural bootstrapping refers to mechanisms relying on prior knowledge, sensorimotor experience, and inference that can be implemented in robotic systems and employed to speed up learning and problem solving in new environments. In this context, the interplay between the symbolic encoding of the sensorimotor information, prior knowledge, planning, and natural language understanding plays a significant role. In this paper, we show how the symbolic descriptions of the world can be generated on the fly from the continuous robot's memory. We also introduce a multi-purpose natural language understanding framework that processes human spoken utterances and generates planner goals as well as symbolic descriptions of the world and human actions. Both components were tested on the humanoid robot ARMAR-III in a scenario requiring planning and plan recognition based on human-robot communication.

Natural language understanding method and natural language understanding system

Abstract: The invention discloses a natural language understanding method and a natural language understanding system. The method includes the following steps: a sequencing model is constructed in advance; natural language information to be parsed is semantically parsed, so that a plurality of semantic parsing results are obtained; the correlation characteristics of the semantic parsing results are extracted; the correlation characteristics are inputted into the sequencing model, so that the sequencing score of each semantic parsing result outputted by the sequencing model is obtained; according to the sequencing scores, one or more of the semantic parsing results are chosen as natural language understanding results. The invention can be utilized to effectively increase the reliability and accuracy of natural language understanding.

Poster: ProNat: An Agent-Based System Design for Programming in Spoken Natural Language

Abstract: The emergence of natural language interfaces has led to first attempts of programming in natural language. We present ProNat, a tool for script-like programming in spoken natural language (SNL). Its agent-based architecture unifies deep natural language understanding (NLU) with modular software design. ProNat focuses on the extraction of processing flows and control structures from spoken utterances. For evaluation we have begun to build a speech corpus. First experiments are conducted in the domain of domestic robotics, but ProNat's architecture makes domain acquisition easy. Test results with spoken utterances in ProNat seem promising, but much work has to be done to achieve deep NLU.

Nlu training with merged engine and user annotations

Abstract: Techniques for training a natural language understanding (NLU) engine may include generating a first annotation of free-form text documenting a healthcare patient encounter and a link between the first annotation and a corresponding portion of the text, using the NLU engine. A second annotation of the text and a link between the second annotation and a corresponding portion of the text may be received from a human user. The first annotation and its corresponding link may be merged with the second annotation and its corresponding link. Training data may be provided to the engine in the form of the text and the merged annotations and links.

Narrative hermeneutic circle: improving character role identification from natural language text via feedback loops

Abstract: While most natural language understanding systems rely on a pipeline-based architecture, certain human text interpretation methods are based on a cyclic process between the whole text and its parts: the hermeneutic circle. In the task of automatically identifying characters and their narrative roles, we propose a feedback-loop-based approach where the output of later modules of the pipeline is fed back to earlier ones. We analyze this approach using a corpus of 21 Russian folktales. Initial results show that feeding back high-level narrative information improves the performance of some NLP tasks.

On Z-numbers and the Machine-Mind for Natural Language Comprehension

Abstract: This article is centred on two themes. The first is the extension of Zadeh’s basic Z-numbers into a tool for level-2 Computing With Words (CWW) and consequently subjective natural language understanding. We describe an algorithm and new operators (leading to complex or spectral Z-numbers), use them to simulate differential diagnosis, and highlight the inherent strengths and challenges of the Z-numbers. The second theme deals with the design of a, Minsky’s Society of Mind based, natural language comprehending machine-mind architecture. We enumerate its macro-components (function modules and memory units) and illustrate its working mechanism through simulation of metaphor understanding; validating system outputs against human-comprehension responses. The framework uses the aforementioned new Z-number paradigm to precisiate knowledge-frames. The completeness of the conceptualized architecture is analyzed through its coverage of mind-layers (Minsky) and cerebral cortex regions. The research described here draws from multiple disciplines and seeks to contribute to the cognitive-systems design initiatives for man-machine symbiosis.

A Corpus of Rich Metaphor Annotation

Abstract: Metaphor is a central phenomenon of language, and thus a central problem for natural language understanding. Previous work on the analysis of metaphors has identified which target concepts are being thought of and described in terms of which source concepts, but this is not adequate to explain what motivates the use of particular metaphors. This work proposes the use of conceptual schemas to represent the underspecified scenarios that motivate a metaphoric mapping. To support the creation of systems that can understand metaphors in this way, we have created and are publicly releasing a corpus of manually validated metaphor annotations.

Personalizing Natural Language Understanding Systems

Abstract: Examples of the present disclosure describe systems and methods of personalizing natural language systems. In aspects, personal data may be uploaded to a personalization server. Upon receiving a data request, a server device may query the personalization server using a user's login information. The login data and the associated personal data may be paired and provided to the personalization server. The paired data may then be provided to a language understanding model to generate a response to the data request. The data in the response may be used to train the language understanding model.

Hybrid Approach to Pronominal Anaphora Resolution in English Newspaper Text

Abstract: One of the challenges in natural language understanding is to determine which entities to be referred in the discourse and how they relate to each other. Anaphora resolution needs to be addressed in almost every application dealing with natural language such as language understanding and processing, dialogue system, system for machine translation, discourse modeling, information extraction. This paper represents a system that uses the combination of constraint- based and preferences-based architectures; each uses a different source of knowledge and proves effective on computational and theoretical basis, instead of using a monolithic architecture for anaphora resolution. This system identifies both inter-sentential and intra-sentential antecedents of "Third person pronoun anaphors" and "Pleonastic it". This system uses Charniak Parser (parser05Aug16) as an associated tool, and it relays on the output generated by it. Salience measures derived from parse tree are used in order to find out accurate antecedents from the list of all potential antecedents. We have tested the system extensively on 'Reuters Newspaper corpus' and efficiency of the system is found to be 81.9%.

Natural language understanding cache

Abstract: Disclosed methods and systems are directed to natural language understanding cache usage. The methods and systems may include receiving a first natural language input comprising a set of one or more terms, and parsing the first natural language input to determine a first pretag result comprising a first string comprising at least one term from the set of one or more terms. The methods and systems may also determine that if the first pretag result corresponds to a key stored in a cache, then retrieve one or more cached NLU results corresponding to the at least one key. The methods and systems may also determine that if the first pretag result does not correspond to a key stored in the cache, then determine, based on the set of one or more terms, a first NLU result corresponding to the first natural language input.

Natural Language Direction Following for Robots in Unstructured Unknown Environments

Abstract: : Robots are increasingly performing collaborative tasks with people in homes, workplaces, and outdoors, and with this increase in interaction comes a need for e cient communication between human and robot teammates. One way to achieve this communication is through natural language, which provides a exible and intuitive way to issue commands to robots without requiring specialized interfaces or extensive user training. One task where natural language understanding could facilitate humanrobot interaction is navigation through unknown environments, where a user directs a robot toward a goal by describing \201in natural language\202 the actions necessary to reach the destination. Most existing approaches to following natural language directions assume that the robot has access to a complete map of the environment ahead of time. This assumption severely limits the potential environments in which a robot could operate, since collecting a semantically labeled map of the environment is expensive and time consuming. Following directions in unknown environments is much more challenging, as the robot must now make decisions using only information about the parts of the environment it has observed so far. In other words, absent a full map the robot must incrementally build up its map \201using sensor measurements\202, and rely on this partial map to follow the direction. Some approaches to following directions in unknown environments do exist, but they implicitly restrict the structure of the environment, and have so far only been applied in simulated or highly structured environments. To date, no solution exists to the problem of real robots following natural directions through unstructured and unknown environments. We address this gap by formulating the problem of following directions in unstructured unknown environments as one of sequential decision making under uncertainty. In this setting, a policy reasons about the robot's knowledge of the world so far, and predicts a sequ

Use of background knowledge in natural language understanding for information fusion

Abstract: Tractor is a system for understanding English messages within the context of hard and soft information fusion for situation assessment. Tractor processes a message through text processors, and stores the result, expressed in a formal knowledge representation language, in a syntactic knowledge base. This knowledge base is enhanced with ontological and geographic information. Finally, Tractor applies hand-crafted syntax-semantics mapping rules to convert the enhanced syntactic knowledge base into a semantic knowledge base containing the information from the message enhanced with relevant background information. Throughout its processing, Tractor makes use of various kinds of background knowledge: knowledge of English usage; world knowledge; domain knowledge; and axiomatic knowledge. In this paper, we discuss the various kinds of background knowledge Tractor uses, and the roles they play in Tractor's understanding of the messages.

Exploiting online human knowledge in Requirements Engineering

Abstract: Data-driven Natural Language Processing (NLP) methods have noticeably advanced in the past few years. These advances can be tied to the drastic growth of the quality of collaborative knowledge bases (KB) available on the World Wide Web. Such KBs contain vast amounts of up-to-date structured human knowledge and common sense data that can be exploited by NLP methods to discover otherwise-unseen semantic dimensions in text, aiding in tasks related to natural language understanding, classification, and retrieval. Motivated by these observations, we describe our research agenda for exploiting online human knowledge in Requirements Engineering (RE). The underlying assumption is that requirements are a product of the human domain knowledge that is expressed mainly in natural language. In particular, our research is focused on methods that exploit the online encyclopedia Wikipedia as a textual corpus. Wikipedia provides access to a massive number of real-world concepts organized in hierarchical semantic structures. Such knowledge can be analyzed to provide automated support for several exhaustive RE activities including requirements elicitation, understanding, modeling, traceability, and reuse, across multiple application domains. This paper describes our preliminary findings in this domain, current state of research, and prospects of our future work.