Predicate Representation Research Articles

Decomposed Prototype Learning for Few-Shot Scene Graph Generation

Today's scene graph generation (SGG) models typically require abundant manual annotations to learn new predicate types. Therefore, it is difficult to apply them to real-world applications with massive uncommon predicate categories whose annotations are hard to collect. In this paper, we focus on Few-Shot SGG (FSSGG) , which encourages SGG models to be able to quickly transfer previous knowledge and recognize unseen predicates well with only a few examples. However, current methods for FSSGG are hindered by the high intra-class variance of predicate categories in SGG: On one hand, each predicate category commonly has multiple semantic meanings under different contexts. On the other hand, the visual appearance of relation triplets with the same predicate differs greatly under different subject-object compositions. Such great variance of inputs makes it hard to learn generalizable representation for each predicate category with current few-shot learning (FSL) methods. However, we found that this intra-class variance of predicates is highly related to the composed subjects and objects. To model the intra-class variance of predicates with subject-object context, we propose a novel Decomposed Prototype Learning (DPL) model for FSSGG. Specifically, we first construct a decomposable prototype space to capture diverse semantics and visual patterns of subjects and objects for predicates by decomposing them into multiple prototypes. Afterwards, we integrate these prototypes with different weights to generate query-adaptive predicate representation with more reliable semantics for each query sample. We conduct extensive experiments and compare with various baseline methods to show the effectiveness of our method.

Requirements prioritization for complex products based on fuzzy associative predicate representation learning

Complex products, integrating many structures and functions, face challenges in fulfilling all requirements due to limited time and resources, making Requirements Prioritization (RP) essential in product development. The complexity of RP increases with the need to consider a wider range of criteria and data from more stakeholders like developers and customers, introducing uncertainty in requirements and expert opinions. However, current research rarely explores systematic methods for addressing requirements in this uncertain environment. Based on that, this paper presents a hybrid framework for organizing knowledge related to RP and determining item priorities. Specifically, we build a multi-dimensional evaluation indicator ontology, model requirement knowledge based on fuzzy RDF Knowledge Graph(KG), generate fuzzy membership degree through representation learning, and then rank requirements by fuzzy soft set. Finally, the effectiveness of our framework is validated in two aspects: evaluation of the fuzzy associative predicate representation learning method and application through a practical case study.

Adaptive Feature Learning for Unbiased Scene Graph Generation.

Scene Graph Generation (SGG) aims to detect all objects and identify their pairwise relationships in the scene. Recently, tremendous progress has been made in exploring better context relationship representations. Previous work mainly focuses on contextual information aggregation and uses de-biasing strategies on samples to eliminate the preference for head predicates. However, there remain challenges caused by indeterminate feature training. Overlooking the label confusion problem in feature training easily results in a messy feature distribution among the confused categories, thereby affecting the prediction of predicates. To alleviate the aforementioned problem, in this paper, we focus on enhancing predicate representation learning. Firstly, we propose a novel Adaptive Message Passing (AMP) network to dynamically conduct information propagation among neighbors. AMP provides discriminating representations for neighbor nodes under the view of de-noising and adaptive aggregation. Furthermore, we construct a feature-assisted training paradigm alongside the predicate classification branch, guiding predicate feature learning to the corresponding feature space. Moreover, to alleviate biased prediction caused by the long-tailed class distribution and the interference of confused labels, we design a Bi-level Curriculum learning scheme (BiC). The BiC separately considers the training from the feature learning and de-biasing levels, preserving discriminating representations of different predicates while resisting biased predictions. Results on multiple SGG datasets show that our proposed method AMP-BiC has superior comprehensive performance, demonstrating its effectiveness.

REUSE OF INFORMATION BASED ON THE INTERPRETATION OF KNOWLEDGE

Recently, much attention has been paid to the creation of knowledge bases that contain millions of facts about various objects of the real world. One of the key aspects of knowledge management is the reuse of previously acquired knowledge. The subject of research is the processes of knowledge reuse and the creation of software systems based on knowledge bases. Knowledge interpretation is one approach to knowledge reuse, which consists in deriving new knowledge based on existing facts in the knowledge base. The purpose of the study is to increase the efficiency of knowledge reuse in software systems based on knowledge bases due to automatic rule extraction. To achieve the goal, the following tasks were solved: approaches to structuring the facts available in the database were considered, a qualitative analysis of the possibility of applying automatic methods of rule construction and derivation was carried out. The task of predicting the connection between a pair of entities, which determines the presence of a relationship for facts, is considered. A generalized approach to the presentation of facts is proposed, which allows the use of efficient rule-finding algorithms. The following methods are used to solve the given problem: the algebra of finite predicates and predicate operations for knowledge representation, methods for predicting the connection between a pair of entities based on representative learning for automatically obtaining rules. The following results were obtained: an approach to rule formation was considered, which allows structuring existing facts as a set of binary predicates and applying automatic methods of rule construction and derivation. It is concluded that the limitation of knowledge reuse is the structure of the knowledge base and the software used to support it. The article formulates the principles of building specific concentrator predicates for the representation of attributes, which allows generalizing the predicate representation of facts and applying automatic methods of rule extraction, which increases the efficiency of knowledge reuse. Conclusions: the application of the method and mechanism of identification based on predicate operations and specific predicates, which automatically extracts attributes from the knowledge base, together with the quality assessment of the derived rules, made it possible to propose a generalized approach for presenting facts and use effective rule search algorithms, which allows to increase the efficiency of reuse knowledge in software systems.

On improving knowledge graph facilitated simple question answering system

Leveraging knowledge graph will benefit question answering tasks, as KG contains well-structured informative data. However, training knowledge graph-based simple question answering systems is known computationally expensive due to the complex predicate extraction and candidate pool generation. Moreover, the existing methods based on convolutional neural network (CNN) or recurrent neural network (RNN) overestimate the importance of predicate features thus reduce performance. To address these challenges, we propose a time-efficient and resource-effective framework. We use leaky n-gram to balance recall and candidate pool size in candidate pool generation. For predicate extraction, we propose a soft-histogram and self-attention (SHSA) module which serves the role of preserving the global information of questions via feature matrices. And this leads to reduce the RNN module as the simple feedforward network in predicate representation. We also designed a Hamming lower-bound label encoding algorithm to encode the label representations in lower dimensions. Experiments on benchmark datasets show that our method outperforms the competitive work for end-tasks and achieves better recall with a significantly pruned candidate space.

Scene Graph Generation With Hierarchical Context.

Scene graph generation has received increasing attention in recent years. Enhancing the predicate representations is an important entry point to this task. There are various methods to fully investigate the context of representation enhancement. In this brief, we analyze the decisive factors that can significantly affect the relation detection results. Our analysis shows that spatial correlations between objects, focused regions of objects, and global hints related to the relations have strong influences in relation prediction and contradiction elimination. Based on our analysis, we propose a hierarchical context network (HCNet) to generate a scene graph. HCNet consists of three contexts, including interaction context, depression context, and global context, which integrates information from pair, object, and graph levels. The experiments show that our method outperforms the state-of-the-art methods on the Visual Genome (VG) data set.

On bijunctive predicates over a finite set

Abstract The paper is concerned with representations of predicates over a finite set in the form of generalized conjunctive normal forms (GCNF). Properties of predicates GCNF are found which are preserved by some majority function. Such predicates are called generalized bijunctive predicates. These properties are used to construct new faster polynomial algorithms for the generalized satisfiability problem in the case when some majority function preserves all the original predicates.

Mining predicate-based entailment rules using deep contextual architecture

Abstract Semantic inference plays an essential role in numerous Natural Language Processing (NLP) tasks, such as question answering, machine reading and text summarization. Reasoning in natural language is inseparable from the knowledge about inference, which is often represented in the form of predicate-based entailment rules. Many efforts have been dedicated to extracting entailment rules from text corpora by utilizing statistical methodology including distributional hypothesis and Latent Dirichlet Allocation (LDA). However, these studies could not give equal consideration to both coverage and accuracy of the mined rules, which brings instability to downstream applications. To solve this problem, this paper proposes a novel model named Deep Contextual Architecture (DCA), which is driven by Deep Belief Networks (DBNs), for the task of mining predicate-based inference rules from texts. Besides previously used statistical contextual information, we also involve semantic meanings represented by word embeddings into DBNs to learn topic level representation of predicates. Combining benefits from both kinds of information, the proposed DCA model shows potential for better modeling the context of predicates. Evaluation on public datasets demonstrates that our method outperforms several strong baselines.

A Categorical Foundation for Structured Reversible Flowchart Languages

Structured reversible flowchart languages is a class of imperative reversible programming languages allowing for a simple diagrammatic representation of control flow built from a limited set of control flow structures, as ordinary structured flowcharts allow for conventional languages. This class includes the reversible programming language Janus (without recursion), as well as more recently developed reversible programming languages such as R-CORE and R-WHILE. In the present paper, we develop a categorical foundation for this class of languages based on inverse categories with joins. We generalize the notion of extensivity of restriction categories to one that may be accommodated by inverse categories, and use the resulting decision maps to give a reversible representation of predicates and assertions. This leads to a categorical semantics for structured reversible flowcharts, from which we show that a program inverter can be extracted. Finally, we exemplify our approach by the development of a small structured reversible flowchart language, use our framework to both straightforwardly give it semantics and derive fundamental theorems about it, and discuss further applications of decisions in reversible programming.

Albańskie odczasownikowe derywaty przymiotnikowe

Albanian adjective derivatives derived from verbsThe aim of this article is the analysis of the Albanian adjective derivatives which are derived from verbs. The main assumption made by the author is the existence of structural isomorphism (parallelism in language elements system in which structuring of one level parallels or is made to parallel that of another) between the structure of the derivative and the structure of the sentence. The proposed model of the analysis is directed from content to form. All examined adjective derivatives derived from verbs are being analyzed with their noun context which displays their semantic structure. Nominal phrases containing adjective derivatives are interpreted as the representation of predicate–argument structures. The kind of relation between defined noun and derivational base of the adjective is a criterion of the division of Albanian adjective derivatives derived from verbs. These main types of relations are:relation in which the base of an adjective represents basic predicate;relation in which the base of an adjective represents additional predicate.Classes which are the results of the main division are the subject to detailed description.

The syntax of argument structure by Leonard H. Babby (review)

Reviewed by: The syntax of argument structure by Leonard H. Babby Peter Arkadiev Leonard H. Babby. The syntax of argument structure. Cambridge: Cambridge University Press, 2009, pp. xviii + 307. [Cambridge Studies in Linguistics, 120.] In the book under review, Leonard Babby,1 an eminent linguist renowned for his influential work on many aspects of the syntax of Russian, presents an ingenious and sophisticated theory of morphosyntax stemming from his previous studies and in many important ways diverging from the current mainstream of generative grammar. While for most proponents of Government and Binding and Minimalist Program “argument structure” is either read off the semantic representation of predicates (as in, e.g., Levin and Rappaport Hovav 1995as in, e.g., Levin and Rappaport Hovav 1998) or reduced to syntactic configurations (e.g., Borer 2005, Ramchand 2008), Babby argues for a view that argument structure is an autonomous module of grammar, and in fact the core one, from which syntactic representations are built, and which cannot be fully reduced to semantics. For Babby, argument structure is not just a static representation, but a dynamic component of language where various operations such as passivization, causativization, etc., apply, and which is crucial for the determination of the initial syntactic representation of clauses. Despite its explicit theory-constructing goals, the book is thoroughly empirically-oriented, presenting a number of in-depth case studies of Russian morphosyntax, such as the structure and syntax of “short” (SF) and “long” (LF) forms of adjectives, control of infinitives, participles and gerunds, and the predicate instrumental construction, each elucidating particular aspects of the theory. A reader, accustomed to associating the term “argument structure” with such issues as “unaccusativity,” “dative shift,” or “spray/load alternations,” might be surprised by Babby’s choice of subject matter. However, the book shows, and quite convincingly, [End Page 259] that the phenomena Babby investigates can be elegantly and revealingly accounted for from the chosen perspective. The book consists of an introduction, five chapters, endnotes, references and indices. In the introduction (pp. 1–10) Babby provides the reader with a very useful nutshell presentation of the main theoretical claims of the book and locates his conception of argument structure against the background of current generative theorizing. Importantly, Babby acknowledges from the outset that his theoretical thinking has been largely motivated by his empirical work on Russian, whose morphosyntax, being different from English in many important respects, cannot necessarily be adequately accounted for by “Anglocentric” approaches reducing all grammatical complexity to basic syntactic operations such as Merge and Move. In chapter 1 “The structure of argument structure” (pp. 11–73) Babby presents the conceptual and technical basics of his theory of argument structure and syntactic projection. The theory rests on the notion of diathesis consisting of two “autonomous but related tiers” (p. 13) of theta selection and category selection, in part going back to the corresponding notion developed in the Leningrad school of typology (Xolodovič 1974). Two assumptions about diathesis underlying the whole theory are the following: (i) all predicates and productive predicate affixes in all languages have the same universal diathesis structure (see below); (ii) syntactic operations such as movement cannot alter the predicate’s diathesis and, consequently, the basic grammatical relations in the clause. From (ii) it follows that many argument-structure-related operations often assumed to be syntactic, e.g., passivization or subject suppression with infinitives, occur at the level of argument structure and not in syntax proper. Therefore, argument structure is a dynamic morphosyntactic component of grammar where the lexically determined diathesis of the verbal stem is successively combined (“amalgamated”) with the diatheses of whatever productive affixes2 the verb attaches, and this interaction between the diatheses of the predicate stem and affixes determines the potential syntactic configurations in which the predicate occurs. Technically, the diathesis is a 2 × 4 grid where the upper tier represents up to three thematic roles and the lower tier, in turn, the syntactic categories corresponding to them; see (1) below for a tabular notation [End Page 260] and (2) for the corresponding linear notation of the diathesis of a three-argument ditransitive predicate. The order of theta positions in the upper tier and their mapping to phrase structure...

Synthesis of a Class of Discrete-Event Controllers for Large Manufacturing Systems

A synthesis method is presented for a class of discrete-event controllers employed in large manufacturing systems forcing at most one control action in each stage of the control pattern. The theoretical grounds are provided by a predicate representation of a synthesis framework based on supervisory control theory for arbitrary control patterns. Algorithms are established for synthesizing complete nonblocking controllers. These algorithms are part of a synthesis procedure deployed in an experimental tool using binary decision diagrams. Two examples show how to use the proposed method and its associated tool, highlighting their benefits and disadvantages. The examples also illustrate how to apply the method in order to maintain the modeling simple and to synthesize controllers for manufacturing systems of considerable size that otherwise could not be obtained.

Horn classes of predicate systems and varieties of partial algebras

We propound an approach through which techniques of the theory of quasivarieties of predicate systems are brought to bear on partial algebras. For every partial algebra A, two predicate representations are treated. The first is the graph of A whose basic operations are graphs of the basic operations on A. The second representation results from the graph of A by adding domains of the operations on A to its basic relations. Studying partial algebras from various perspectives makes it necessary to deal with different equality semantics. Here we present a general definition of semantics that stretches over such instances as weak semantics, Evans’ semantics. Kleene semantics, and strong semantics. On a set of all semantics, the preorder is induced in increasing “force,” and it is proved that certain of the properties of varieties of partial algebras in a given semantics are individuated by the position it takes in that set. We argue that every variety of partial algebras, in any semantics, is in correspondence with a Horn class of predicate systems which admits a generation operator and is closed under direct limits and retracts. For such classes we prove analogs of the Birkhoff theorem on subdirect decompositions and of the Taylor theorem on residual smallness. Therefore, these are also applicable to varieties of partial algebras in arbitrary semantics.

Cognitive representations of predicates and the use of past tenses in French: a developmental approach

Two experiments examine how French 10-year-old children and adults relate past tenses to the semantic properties of predicate types in writing. Experiment 1 involved two tasks: graphically coding two predicate dimensions (durativity and resultativity), a task designed to assess cognitive representations of these predicates; and selecting past tenses in sentences which included previously evaluated verbs. Results show that (a) 10-year-olds and adults have comparable representations of durativity, but different ones for resultativity, (b) the adults associate process charac teristics and past tenses, but 10-year-olds do not. In Experiment 2, subjects only had to code graphically two dimensions of the predicates (durativity and resultativity) in sentences where the verb could take five forms: infinitive, one imperfective past tense (imperfect, 'imparfait'), and three perfective past tenses (perfect 'passé composé', or its 'historic' form 'passé simple', and pluperfect 'plus-que-parfait'). Results show that 10-year-old children have not yet learned the 'adult' functions of verbal forms. The results of these two experiments are discussed in relation to current models of human cognition.

A machine learning approach for acquiring descriptive classification rules of shape contours

We devise a method to generate descriptive classification rules of shape contours by using inductive learning. The classification rules are represented in the form of logic programs. We first transform input objects from pixel representation into predicate representation. The transformation consists of preprocessing, feature extraction and symbolic transformation. We then use FOIL which is an indictive logic programming system to produce classification rules. Experiments on two sets of data were performed to justify our proposed method.

Cognitive representations of predicates and the use of past tenses in French: a developmental approach

Two experiments examine how French 10-year-old children and adults relate past tenses to the semantic properties of predicate types in writing. Experiment 1 involved two tasks: graphically coding two predicate dimensions (durativity and resultativity), a task designed to assess cognitive representations of these predicates; and selecting past tenses in sentences which included previously evaluated verbs. Results show that (a) 10-year-olds and adults have comparable representations of durativity, but different ones for resultativity, (b) the adults associate process charac teristics and past tenses, but 10-year-olds do not. In Experiment 2, subjects only had to code graphically two dimensions of the predicates (durativity and resultativity) in sentences where the verb could take five forms: infinitive, one imperfective past tense (imperfect, 'imparfait'), and three perfective past tenses (perfect 'passé composé', or its 'historic' form 'passé simple', and pluperfect 'plus-que-parfait'). Results show that 10-year-old children have not yet learned the 'adult' functions of verbal forms. The results of these two experiments are discussed in relation to current models of human cognition.

SILOL: A simple logical-linguistic document retrieval system

This paper introduces a logical-linguistic model of document retrieval systems and describes an implementation of a system called SILOL which is based on this model. SILOL uses a shallow semantic translation of natural language texts into a first order predicate representation in performing a document indexing and retrieval process. Some preliminary experiments have been carried out to test the retrieval effectiveness of this system. The results obtained show improvements in the level of retrieval effectiveness, which demonstrate that the approach of using a semantic theory of natural language and logic in document retrieval systems is a valid one.