SPARQA: Skeleton-Based Semantic Parsing for Complex Questions over Knowledge Bases

Abstract

Semantic parsing transforms a natural language question into a formal query over a knowledge base. Many existing methods rely on syntactic parsing like dependencies. However, the accuracy of producing such expressive formalisms is not satisfying on long complex questions. In this paper, we propose a novel skeleton grammar to represent the high-level structure of a complex question. This dedicated coarse-grained formalism with a BERT-based parsing algorithm helps to improve the accuracy of the downstream fine-grained semantic parsing. Besides, to align the structure of a question with the structure of a knowledge base, our multi-strategy method combines sentence-level and word-level semantics. Our approach shows promising performance on several datasets.

SemML: Facilitating Development of ML Models for Condition Monitoring with Semantics

“…Each qmm-ml:FeatureProcessingLayer or qmmml:MLModellingLayer has a structure of qmm-ml:Input, qmm-ml: Algorithm, and qmm-ml:Output. A piece of the ML pipeline ontology looks like this: (8) Individual: qmm-ml-i:p1 (9) Types: qmm-ml:Pipeline (10) Facts: qmm-ml:hasNextLayer qmm-ml-i:l1 (11) Individual: qmm-ml-i:l1 (12) Types: qmm-ml:FeatureProcessingLayer (13) Facts: qmm-ml:hasNextLayer qmm-ml-i:l2, :hasInputOutputCombination io1, io2 (14) Individual: qmm-ml-i:io1 (15) Types: qmm-ml:InputOutputCombination (16) Facts: qmm-ml:hasInput :FG-SingleFeature, qmm-ml:hasAlgorithm qmm-ml:Maintain, qmm-ml:hasOutput :FG-SingleFeatureMaintained (17) Individual: qmm-ml-i:io2 (18) Types: qmm-ml:InputOutputCombination (19) Facts: qmm-ml:hasInput :FG-ProcessCurve, qmm-ml:hasAlgorithm qmm-ml:GetStats, qmm-ml:hasOutput qmm-ml:FG-TSStats…”

“…We now illustrate this reasoning with an example: (18) Class: qmm-rsw:ElectrodeCap (19) SubClassOf: qmm-core:SystemComponent (20) SubClassOf: qmm-rsw:hasElectrodeCapStatus only qmm-rsw:WearCount (21) Class: qmm-rsw:WearCount (22) SubClassOf: qmm-core:ToolWearingStatus (23) Class: qmm-core:ToolWearingStatus (24) SubClassOf: qmm-core:Status (25) SubClassOf: qmm-core:hasMLFeatureGroup only qmm-ml:FG-Wear Axiom 20 defines that (the elements of the class) qmm-rsw: ElectrodeCap can only have the status parameter qmm-rsw: WearCount. The latter has a superclass qmm-core:ToolWearingStatus from the Core ontology (Axiom 22), which is assigned the only associated machine learning feature group qmm-ml:FG-Wear (axiom 25).…”

“…In this work we address the C1-C3 challenges by enhancing machine learning development for condition monitoring with semantic technologies. Note that semantics has recently gained a considerable attention in industry in a wide range of applications and automation tasks such as modelling of industrial assets [12] industrial analytics [13], integration [14][15][16] and querying [17][18][19] of production data, process monitoring [20] and equipment diagnostics [21], moreover, semantic technologies have been adopted or evaluated in a number of large high tech production companies such as Equinor [22], Siemens [23], Festo [24], and Bosch [25,26].…”

Injecting Numerical Reasoning Skills into Knowledge Base Question Answering Models

“…The task has attracted much attention because logically organized entities and relations can explicitly promote the answer reasoning process. Recently, researchers focus more on the complex KBQA task which involves multi-hop reasoning, constrained reasoning, and numerical reasoning (Bao et al, 2016;Lan and Jiang, 2020;Sun et al, 2020;Kapanipathi et al, 2021). Existing methods for KBQA tasks can be generally categorized into semantic parsing-based (SP-based) methods (Lan and Jiang, 2020;Sun et al, 2020) and embedding-based methods (Sun et al, 2018;Saxena et al, 2020;.…”

Dual‐Channel Reasoning Model for Complex Question Answering

“…[17] proposed a modified staged query graph generation method by allowing longer relation paths. Sun et al [18] proposed a novel skeleton grammar that uses the BERT-based parsing algorithm to improve the downstream fine-semantic parsing. To avoid generating noisy candidate queries, Chen et al [19] proposed abstract query graphs (AQG) to describe query structures.…”