Filename
stringlengths 22
64
| Paragraph
stringlengths 8
5.57k
|
|---|---|
Processed_Second_moment_of_the_pion's_distribution_amplitude.txt | Using a fourth order Gegenbauer polynomial expansion, we calculate a value for the sec- ond Gegenbauer moment, a2(µ2 = 5 GeV2) = 0.236(82). |
Processed_Second_moment_of_the_pion's_distribution_amplitude.txt | Although we have only employed a linear chi- ral extrapolation and our operators are not O(a)- improved, the chiral and continuum extrapola- tions do not seem to be a major source of sys- tematic error when compared to the statistical errors. These issues will be addressed in more de- tail in a forthcoming coming publication, where we also intend to investigate finite size and (par- tially) quenching effects as well as renormalisation group running of the relevant matrix elements. |
Processed_Second_moment_of_the_pion's_distribution_amplitude.txt | The numerical calculations have been done on the Hitachi SR8000 at LRZ (Munich), on the Cray T3E at EPCC (Edinburgh) [23] and on the APE1000 at DESY (Zeuthen). This work was supported in part by the DFG (Forschergruppe Gitter-Hadronen-Ph¨anomenologie) and by the EU Integrated Infrastructure Initiative Hadron Physics (I3HP) under contract RII3-CT-2004- 506078. |
Processed_One-Shot_Affordance_Detection.txt | Affordance detection refers to identifying the po- tential action possibilities of objects in an image, which is an important ability for robot percep- tion and manipulation. To empower robots with this ability in unseen scenarios, we consider the challenging one-shot affordance detection prob- lem in this paper, i.e., given a support image that depicts the action purpose, all objects in a scene with the common affordance should be detected. To this end, we devise a One-Shot Affordance Detection (OS-AD) network that firstly estimates the purpose and then transfers it to help detect the common affordance from all candidate images. Through collaboration learning, OS-AD can cap- ture the common characteristics between objects having the same underlying affordance and learn a good adaptation capability for perceiving unseen affordances. Besides, we build a Purpose-driven Affordance Dataset (PAD) by collecting and label- ing 4k images from 31 affordance and 72 object categories. Experimental results demonstrate the superiority of our model over previous representa- tive ones in terms of both objective metrics and vi- sual quality. The benchmark suite is at ProjectPage. |
Processed_One-Shot_Affordance_Detection.txt | Figure 1: Illustration of perceiving affordance. Given a support im- age that depicts the action purpose, all objects in a scene with the common affordance could be detected. |
Processed_One-Shot_Affordance_Detection.txt | Figure 2: The part (A) shows that objects usually have multiple affordance. The part (B) shows that the objects with different semantic categories may have the same affordance. |
Processed_One-Shot_Affordance_Detection.txt | tween different objects to counteract the interference caused by visual appearance differences and to improve generaliza- tion. Specifically, we devise a novel One-Shot Affordance Detection (OS-AD) network to solve the problem. Taking an image as support and a set of images (5 images in this pa- per) as a query, the network first captures the human-object interactions from the support image using a purpose learning module (PLM) to encode the action purpose. Then, a pur- pose transfer module (PTM) is devised to use the encoding of the action purpose to activate the features in query images that have the common affordance. Finally, a collaboration en- hancement module (CEM) is devised to capture the intrinsic relationships between objects with the same affordance and suppress backgrounds that are irrelevant to the action pur- pose. In this way, OS-AD can learn a good adaptation ca- pability for perceiving unseen affordances. |
Processed_One-Shot_Affordance_Detection.txt | Moreover, the existing datasets still have gaps relative to real application scenarios, due to the limitation of its diver- sity. The affordance detection for scene understanding and general applications should be able to learn from the human- object interaction when the robot arrives at a new environ- ment and retrieves the objects in the environment, rather than just finding objects with the same categories or similar ap- pearance. To address the limitations of the datasets, we pro- pose the Purpose-driven Affordance Dataset (PAD), which contains 4,002 diverse images covering 31 affordance cate- gories as well as 72 object categories from different scenes. Moreover, we trained several representative models to ad- dress the problem and compare them with our OS-AD model comprehensively in terms of both objective evaluation met- rics and visual quality. Contributions (1) We introduce a new one-shot affordance detection problem along with a benchmark to facilitate the research for empowering robots with the ability to perceive unseen affordances in real-world scenarios. (2) We propose a novel OS-AD network that can efficiently learn the action purpose and use it to detect the common affordance of all objects in a scene via collaboration learning, resulting in a good adaptation capability that can deal with unseen affor- dances. (3) Experiments on the proposed PAD benchmark demonstrate that OS-AD outperforms state-of-the-art models and can serve as a strong baseline for future research. |
Processed_One-Shot_Affordance_Detection.txt | strategy in AI community to perceive action intentions and thereby to infer visual affordance from the image/video of human and objects. Early works mainly attempted to estab- lish and learn an association between the apparent charac- teristics of objects and their affordance for perceiving affor- dance. [Myers et al., 2015] proposed a framework for jointly locating and identifying the affordance of object parts and presented the RGB-D Part Affordance dataset, which is the first pixel-wise labeled affordance dataset. However, the af- fordances of objects do not simply correspond to represen- tational characteristics, which are shifted in response to the state of interactions between objects and humans. Therefore, [Chuang et al., 2018] considered the problem of affordance reasoning in the real world by taking into account both the physical world and the social norms imposed by the soci- ety, and constructed the ADE-Affordance dataset based on ADE20k [Zhou et al., 2017]. Since the change of an object’s affordance state is usually due to the interaction between the human and the object, further research has begun to con- sider human action as a cue for learning affordance. [Fang et al., 2018] used human-object interactions in demonstra- tion videos to predict the affordance regions of static objects via linking human actions to object affordance, and proposed the OPRA dataset for affordance reasoning. Different from the above existing works, our proposed method aims to ex- plore the multiplicity of affordance by a collaborative learn- ing strategy. To a certain extent, our work conforms to Gib- son’s definition of affordance that “it implies the complemen- tarity of the animal and the environment”. Since there exist multiple potential complementarities between animal and en- vironment, it leads to multiple possibilities of particular af- fordance. To address this issue, we present a novel task of one-shot affordance detection, in which action intension is introduced through support image to alleviate the multiplic- ity of affordance. |
Processed_One-Shot_Affordance_Detection.txt | Figure 3: Our One-Shot Affordance Detection (OS-AD) network. OS-AD consists of three key modules: Purpose Learning Module (PLM), Purpose Transfer Module (PTM), and Collaboration Enhancement Module (CEM), which are detailed in Figure 4. |
Processed_Artificial_intelligence_technologies_to_support_re.txt | Executive Summary This document provides literature review for The Responsible Use of Technology-Assisted Research Assessment project that was commissioned by the This study was funded by the four UK higher the Future Research Assessment Programme education (https://www.jisc.ac.uk/future-research-assessment-programme) to assess how technology, in the form of Artificial Intelligence (AI), can help research evaluation in the future, especially for the Research Excellence Framework (REF). Here, AI is essentially software that automates complex tasks. |
Processed_Artificial_intelligence_technologies_to_support_re.txt | The literature review identifies indicators that associate with higher impact or higher quality research from article text (e.g., titles, abstracts, lengths, cited references and readability) or metadata (e.g., the number of authors, international or domestic collaborations, journal impact factors and authors’ h- index). This includes studies that used machine learning techniques to predict citation counts or quality scores for journal articles or conference papers. The literature review also includes evidence about the strength of association between bibliometric indicators and quality score rankings from previous UK Research Assessment Exercises (RAEs) and REFs in different subjects and years and similar evidence from other countries (e.g., Australia and Italy). In support of this, the document also surveys studies that used public datasets of citations, social media indictors or open review texts (e.g., Dimensions, OpenCitations, Altmetric.com and Publons) to help predict the scholarly impact of articles. The results of this part of the literature review were used to inform the experiments using machine learning to predict REF journal article quality scores, as reported in the AI experiments report for this project. |
Processed_Artificial_intelligence_technologies_to_support_re.txt | The literature review also covers technology to automate editorial processes, to provide quality control for papers and reviewers’ suggestions, to match reviewers with articles, and to automatically categorise journal articles into fields. Bias and transparency in technology assisted assessment are also discussed. |
Processed_Artificial_intelligence_technologies_to_support_re.txt | In addition to the analysis of inputs for the AI system to predict REF journal article scores, as discussed in the main report, the following recommendations are made based on the literature review. |
Processed_Artificial_intelligence_technologies_to_support_re.txt | Implement a system to recommend sub-panel members to review outputs. This would likely be based on the ORCIDs of sub-panel members matching their Scopus/Web of Science/Dimensions/etc. profiles, then using text mining to assess the similarity of their outputs with each sub-panel output to be assessed. The text mining might use article titles, abstracts, field classifications and references. |
Processed_Artificial_intelligence_technologies_to_support_re.txt | review of it near the next REF might provide useful insights for both future AI and future human peer review guidelines for sub-panel members. In the next REF, collate information on inter-reviewer agreement rates within sub-panels for outputs scored before cross-checking between reviewers. Use this to assess the human level agreement rates (for all output types) to use as a benchmark for score prediction AI systems. In the tender for bibliometrics and AI for the next REF (if used), mention the importance of accurate classification for bibliometric indicators, including for the percentile system currently used. |
Processed_Artificial_intelligence_technologies_to_support_re.txt | Longer titles associate with more citations: One simple study gathered the 25 most cited and 25 least cited articles from three medical journals (Lancet, BMJ and Journal of Clinical Pathology; total n=150) published in 2005, finding that number of title words positively correlates (rho=0397) with article citations (Jacques & Sebire, 2010). Similarly, a study of 9,031 articles published in 22 medical (e.g., Lancet and JAMA) and multidisciplinary (e.g., Science and Nature) journals in 2005 also found that articles with longer titles either as measured in characters (rho=0266) or in words (rho=o244) received more citations and this positive association was more common in high impact journals (7 of 8 journals) (Habibzadeh & Yadollahie, 2010). |
Processed_Artificial_intelligence_technologies_to_support_re.txt | Insufficient evidence of an association between title length and citations: In contrast to the above, a paper about 302 research articles published in the journal Addictive Behaviors in 2007 found no correlation between the number of title words and citation counts (Rostami, Mohammadpoorasl, & Hajizadeh, 2014). Similarly, no meaningful association was found between title length and citation counts for 1,825 articles published during 1990 to 2002 in five major marketing journals (Stremersch, Verniers & Verhoef, 2007). An investigation of 2,172 open access articles published in six PLoS (Public Library of Science) journals in 2007 also found no significant correlation between title length and citations (Jamali & Nikzad, 2011). |
Processed_Artificial_intelligence_technologies_to_support_re.txt | Shorter titles associate with more citations: In psychology, a study of 258 articles from 40 journals showed that articles with shorter titles received more citations (Subotic & Mukherjee, 2014). Similarly, an investigation of 423 research articles published in 2008 from BioMed Central (BMC) and Public Library of Science (PLoS) journals found a negative association (albeit very weak, r=-0.104) between title length (number of characters) and article citations (Paiva et al., 2012). |
Processed_Artificial_intelligence_technologies_to_support_re.txt | the relationship between article titles lengths and citation counts. These have different problems in that any differences found could be due to journal or topic style variations. |
Processed_Artificial_intelligence_technologies_to_support_re.txt | Time differences in associations between title length and citations: A study of 302,048 Web of Science economics articles from 1956–2012 found that correlations between title length and citation counts were negative between 1956 and 2000 but positive after 2000 (Guo et al. 2018). This result is partly corroborated by another large report on 140,000 highly cited Scopus papers from 2007-2013 (20,000 papers in each year). It found that highly cited articles with shorter titles tend to attract more citations (negative correlation with title length) (Letchford, Moat, & Preis, 2015). Supporting the correlation found in the second half of the Guo et al. investigation, a large study of 1.3 million articles published in 2012 found a weak positive correlation (rho=0.142) between the number of characters in title and citation counts (Haustein, Costas, & Larivière, 2015), although no other large-scale study seems to have reported changes over time in the relationship between title length and citations. Thus, it seems that there may be a general trend for articles with longer titles to be slightly more cited overall now, reversing an earlier trend. This interdisciplinary finding could be a second order effect of disciplinary differences in title lengths and citation rates, however. |
Processed_Artificial_intelligence_technologies_to_support_re.txt | Disciplinary differences in associations between title length and citations: A study of articles published in journals with the highest impact factors during 1996-2005 found that articles with longer titles received fewer citations in both Sociology (n=2,016; r=-0.046) and Applied Physics (n=23,676; r=- 0.089), but the reverse in General Medicine (n= 6,957; r=-0.166) (Van Wesel, Wyatt, & ten Haaf, 2014). For Web of science articles in Biology and Biochemistry (n=16,058) and Social Sciences (n=15,932) from 2000-2009, title length associates with more citations in both subject areas according to both regression models and Spearman correlations (rho=0.021 and 0.014 respectively) but no association was found in Chemistry (n=16,378) (Didegah & Thelwall, 2013b). |
Processed_Artificial_intelligence_technologies_to_support_re.txt | Journal impact differences in associations between title length and citations: One of the largest studies investigating the association between article length and citations used 4.3 million papers in articles published 1995-2004 in 1500 large journals, finding that for highly cited journals, shorter titles tend to be more cited, whereas for the remaining journals, longer titles tend to be more cited (Sienkiewicz & Altmann, 2016). The former result may be due to strict title length restrictions in the most highly cited journals within the highly cited set. |
Processed_Artificial_intelligence_technologies_to_support_re.txt | Despite all the above studies, it is still not fully clear if (a) shorter titles are more common in high impact journals than in other journals or (b) if longer (or shorter) title lengths have a citation advantage in some fields because they increase readership and therefore are more likely to be subsequently cited. The contradictory findings might be related to other factors related to the studied data sets such as a focus on highly cited articles or journals (Subotic & Mukherjee, 2014; Sienkiewicz & Altmann, 2016), different publication years examined because article title lengths have increased over time (Lewison & Hartley, 2005; Guo et al. 2015; Gnewuch & Wohlrabe, 2017), differing document types (e.g., see Soler, 2007) or other factors, such as changing editorial policies regarding title lengths. |
Processed_Artificial_intelligence_technologies_to_support_re.txt | Colons and hyphens in titles associate with more citations: A colon or a hyphen in an article title may help to make it more readable or may help to express more complex information. These may associate with more citations. For instance, an early study of 150 articles from three medical journals (Lancet, BMJ and Journal of Clinical Pathology; total n=150) showed that colons was significantly more frequent in highly cited articles (Jacques & Sebire, 2010). |
Processed_Artificial_intelligence_technologies_to_support_re.txt | Colons and hyphens in titles associate with less citations: Two small studies (n=423 and n=2,172) found that biomedical article titles with colons or hyphens in their titles received fewer citations than titles that did not have these characters (Paiva et al., 2012; Jamali & Nikzad, 2011, respectively). The difference from the above set is presumably the wider collection of articles analysed. |
Processed_Artificial_intelligence_technologies_to_support_re.txt | Question marks in titles associate with more citations: A paper about 312,879 Web of Science articles in economics published between 1980 and 2015 in 430 journals found non-alphanumeric characters in article titles associated with increased their citation impact, with question marks having the strongest apparent effect: 1.64 extra citations (Gnewuch & Wohlrabe, 2017). Similarly, a recent large- scale study of about 2 million Web of Science journal articles and conference proceedings papers (1945 to 2014) in Computer Science also found that articles with titles ending with a question mark (n=5,682) received 16% more citations than articles not asking questions (n=957,837) (Fiala, Král, & Dostal, 2021). |
Processed_Artificial_intelligence_technologies_to_support_re.txt | Non-alphanumeric characters in titles associate with more citations: A large-scale study of 5% of all Web of Science articles from 1999-2008 (n=642,807) found that 68% had at least one out of 29 non- alphanumeric characters in their titles, with hyphens, colons, and commas being the most common. In general, articles with non-alphanumeric characters in their titles had higher field-normalised citation impact than titles with only alphanumeric characters. However, there were disciplinary differences, and this association was positive in Clinical Medicine, negative in Biological Sciences, and no significant association was found in Agriculture and Food Science (Buter and van Raan 2011). |
Processed_Finite_model_reasoning_over_existential_rules.txt | Ontology-based query answering (OBQA) asks whether a Boolean conjunctive query is satisfied by all models of a logical theory consisting of a relational database paired with an ontology. The introduction of existential rules (i.e., Datalog rules extended with existential quantifiers in rule-heads) as a means to specify the ontology gave birth to Datalog+/-, a framework that has received increasing attention in the last decade, with focus also on decidability and finite controllability to support effective reasoning. Five basic decidable fragments have been singled out: linear, weakly-acyclic, guarded, sticky, and shy. Moreover, for all these fragments, except shy, the important property of finite controllability has been proved, ensuring that a query is satisfied by all models of the theory iff it is satisfied by all its finite models. In this paper we complete the picture by demonstrating that finite controllability of OBQA holds also for shy ontologies, and it therefore applies to all basic decidable Datalog+/- classes. To make the demonstration, we devise a general technique to facilitate the process of (dis)proving finite controllability of an arbitrary ontological fragment. |
Processed_Finite_model_reasoning_over_existential_rules.txt | The problem of answering a Boolean query q against a logical theory consisting of an exten- sional database D paired with an ontology Σ is attracting the increasing attention of scientists in various fields of Computer Science, ranging from Artificial Intelligence (Baget et al. 2011; Calvanese et al. 2013; Gottlob et al. 2014) to Database Theory (Bienvenu et al. 2014; Gottlob et al. 2014; Bourhis et al. 2016) and Logic (P´erez-Urbina et al. 2010; B´ar´any et al. 2014; Gottlob et al. 2013). This problem, called ontology-based query answering, for short OBQA (Cal`ı et al. 2009b), is usually stated as D ∪ Σ |= q, and it is equivalent to checking whether q is satisfied by all mod- els of D ∪ Σ according to the standard approach of first-order logics, yielding an open world semantics. |
Processed_Finite_model_reasoning_over_existential_rules.txt | that D ∪ Σ 6|= q implies that there is a finite model M of D ∪ Σ such that M 6|= q. This is usu- ally stated as D ∪ Σ |= q if, and only if, D ∪ Σ |=fin q (where |=fin stands for entailment under finite models), as the “only if” direction is always trivially true. And there are contexts, like in databases (Johnson and Klug 1984; Rosati 2006; B´ar´any et al. 2014), in which reasoning with respect to finite models is preferred. |
Processed_Finite_model_reasoning_over_existential_rules.txt | In this paper we focus on the Datalog± family, which has been introduced with the aim of “closing the gap between the Semantic Web and databases” (Cal`ı et al. 2012) to provide the Web of Data with scalable formalisms that can benefit from existing database technologies. In fact, Datalog± generalizes two well-known subfamilies of Description Logics called EL and DL-Lite, which collect the basic tractable languages for OBQA in the context of the Semantic Web and databases. In particular, we consider ontologies where Σ is a set of existential rules, each of which is a first-order formula ρ of the form ∀X∀Y(φ(X, Y) → ∃Zp(X, Z)), where the body φ(X, Y) of ρ is a conjunction of atoms, and the head p(X, Z) of ρ is a single atom. |
Processed_Finite_model_reasoning_over_existential_rules.txt | The main decidable Datalog± fragments rely on the following five syntactic properties: weak- acyclicity (Fagin et al. 2005), guardedness (Cal`ı et al. 2013), linearity (Cal`ı et al. 2012), sticki- ness (Cal`ı et al. 2010), and shyness (Leone et al. 2012). And these properties underlie the ba- sic classes called weakly-acyclic, guarded, linear, sticky, and shy, respectively. Several vari- ants and combinations of these classes have been defined and studied too (Baget et al. 2010; Kr¨otzsch and Rudolph 2011; Cal`ı et al. 2012; Civili and Rosati 2012; Gottlob et al. 2013), as well as semantic properties subsuming the syntactic ones (Baget et al. 2009; Leone et al. 2012). |
Processed_Finite_model_reasoning_over_existential_rules.txt | The five basic classes above are pairwise uncomparable, except for linear which is strictly con- tained in both guarded and shy, as depicted in Figure 1. Interestingly, both weakly-acyclic and shy strictly contain datalog —the well-known class with rules of the form ∀X∀Y(φ(X, Y) → p(X)), where existential quantification has been dropped. Moreover, sticky strictly contains joinless —the class collecting sets of rules where each body contains no repeated variable. The lat- ter, introduced by Gogacz and Marcinkowski (2013) to prove that sticky is finitely controllable, plays a central role also in this paper. Finally, both linear and joinless strictly contain inclusion- dependencies —the well-known class of relational database dependencies collecting sets of rules with one single body atom and no repeated variable. |
Processed_Finite_model_reasoning_over_existential_rules.txt | ing (namely reasoning over finite models only, here denoted by |=fin ) is required, then the chase is generally uncomplete, unless ontologies are finitely controllable. Hence, proving this property is of utmost importance, especially in those contexts where finite model reasoning is relevant. |
Processed_Finite_model_reasoning_over_existential_rules.txt | Finite controllability of weakly-acyclic comes for free since every ontology here admits a fi- nite universal model, computed by a variant of the chase procedure which goes under the name of restricted chase (Fagin et al. 2005). Conversely, the proof of this property for the subsequent three classes has been a very different matter. Complex, yet intriguing, constructions have been devised for linear (Rosati 2006; B´ar´any et al. 2014), guarded (B´ar´any et al. 2014), and more re- cently for sticky (Gogacz and Marcinkowski 2013). To complete the picture, we have addressed the same problem for shy and get the following positive result, which is the main contribution of the paper. |
Processed_Finite_model_reasoning_over_existential_rules.txt | Theorem 1.1 Under shy ontologies, D ∪ Σ |= q if, and only if, D ∪ Σ |=fin q. |
Processed_Finite_model_reasoning_over_existential_rules.txt | For the proof, we design in Section 3 and exploit in Section 4 a general technique (our second contribution), called canonical rewriting, to facilitate the process of (dis)proving finite control- lability of an arbitrary ontological fragment of existential rules. By exploiting this technique, we can immediately (re)confirm that linear is finitely controllable since inclusion-dependencies is. In addition, we prove (our third contribution) that sticky-join (Cal`ı et al. 2012), generalizing both sticky and linear, is finitely controllable since sticky is. However, differently from linear and sticky-join, the canonical rewriting of a shy ontology —although it is simpler and still a shy ontology— does not immediately fall in any other known class. Therefore, to prove that shy is finitely controllable, we devise three technical tools on top of the canonical rewriting from which we are able to exploit the fact that joinless is finitely controllable. |
Processed_Finite_model_reasoning_over_existential_rules.txt | Basics. Let C, N and V denote pairwise disjoint discrete sets of constants, nulls and vari- ables, respectively. An element t of T = C ∪ N ∪ V is called term. An atom α is a labeled tuple p(t1, . . . ,tm), where p is a predicate symbol, m is the arity of both p anf α, and t1, . . . ,tm are terms. An atom is simple if it contains no repeated term. We denote by pred(α) the predicate symbol p, and by α[i] the i-th term ti of the α. We also consider propositional atoms, which are simple atoms of arity 0 written without brackets. Given two sets A and B of atoms, a homomorphism from A to B is a mapping h : T → T such that c ∈ C implies h(c) = c, and also p(t1, . . . ,tm) ∈ A im- plies p(h(t1), . . . , h(tm)) ∈ B. As usual, we denote by h(A) = {p(h(t1), . . . , h(tm)) : p(t1, . . . ,tm) ∈ A} ⊆ B. An instance I is a discrete set of atoms where each term is either a constant or a null. |
Processed_Finite_model_reasoning_over_existential_rules.txt | Syntax. A database D is a finite null-free instance. An (existential) rule ρ is a first-order formula ∀X∀Y(φ(X, Y) → ∃Zp(X, Z)), where body(ρ) = φ(X, Y) is a conjunction of atoms, and head(ρ) = p(X, Z) is an atom. Constants may occur in ρ. If Z = /0, then ρ is datalog rule. An ontology Σ is a set of rules. For each rule ρ of Σ, we denote by V(ρ) the set of variables appearing in ρ, by EV(ρ) the set of all existential variables of ρ, and by UV(ρ) the set of all universal variables of ρ. A union of Boolean conjunctive query, UBCQ for short, q is a first- order expression of the form ∃Y1ψ1(Y1) ∨ . . . ∨ ∃Ykψk(Yk), where each ψj(Y j) is a conjunction of atoms. Constants may occur also in q. In case k = 1, then q is simply called BCQ. |
Processed_Finite_model_reasoning_over_existential_rules.txt | h′ ⊇ h|X from {head(ρ)} to I. Moreover, I satisfies Σ, denoted by I |= Σ, if I satisfies each rule of Σ. The models of D ∪ Σ, denoted by mods(D, Σ), consist of the set {I : I ⊇ D and I |= Σ}. An instance I satisfies q, written I |= q, if there is a homomorphism from some ψj(Y j) to I. Also, q is true over D ∪ Σ, written D ∪ Σ |= q, if each model of D ∪ Σ satisfies q. |
Processed_Finite_model_reasoning_over_existential_rules.txt | The chase. Consider a logical theory hD, Σi as above. A rule ρ of Σ is applicable to an in- stance I if there is a homomorphism h from body(ρ) to I that maps the existential variables of ρ to different nulls not occurring in I. If so, hρ, hi(I) = I ∪ h(head(ρ)) defines a chase step. The chase procedure (Deutsch et al. 2008) of D ∪ Σ is any sequence I0 = D ⊂ I1 ⊂ . . . ⊂ Im ⊂ . . . of instances obtained by applying exhaustively the rules of Σ in a fair (e.g., breadth-first) fashion in such a way that, for each i > 0, hρ, hi(Ii−1) = Ii defines a chase step for some ρ and h. We call chase(D, Σ) the (possibly infinite) instance Si>0 Ii. Importantly, different chase steps introduce different nulls. This variant of the chase is called oblivious, and defines a family of isomorphic instances, namely any two such instances are equal modulo renaming of nulls. Hence, without loss of generality, it is common practice to consider the oblivious chase as deterministic and its least fixpoint as unique. The restricted version of this procedure imposes a further condi- tion on each chase step: I 6|= h′(head(ρ)), where h′ = h|UV(ρ). Differently from the oblivious one, it defines a family of homomorphically equivalent instances, each generically denoted by rchase(D, Σ). It is well-known that (r)chase(D, Σ) is a universal model of D ∪ Σ, namely for each M ∈ mods(D, Σ), there is a homomorphism from chase(D, Σ) to M. Hence, given a UBCQ q, it holds that (r)chase(D, Σ) |= q if, and only if, D ∪ Σ |= q (Fagin et al. 2005). |
Processed_Finite_model_reasoning_over_existential_rules.txt | Finite controllability. The finite models of a theory D ∪ Σ, denoted by fmods(D, Σ), are the finite instances in {I ∈ mods(D, Σ) : |I| ∈ N}. An ontological fragment F is finitely controllable if, for each database D, for each ontology Σ of F , and for each UBCQ q, it holds that D ∪ Σ 6|= q implies that there exists a finite model M of D ∪ Σ such that M 6|= q. This is formally stated as D ∪ Σ |= q if and only if D ∪ Σ |=fin q, or equivalently chase(D, Σ) |= q if and only if D ∪ Σ |=fin q. |
Processed_Finite_model_reasoning_over_existential_rules.txt | Fix a database D, an ontology Σ, and a chase step involving some pair h ¯ρ, hi. To lighten the pre- sentation, we assume that different rules of Σ share no variable. Also, for every m-ary predicate p and every i ∈ {1, . . . , m}, the pair (p, i) is called position and denoted by p[i]. Finally, given a set A of atoms, a term t occurs in A at position p[i] if there is α ∈ A s.t. pred(α) = p and α[i] = t. Local conditions. Σ belongs to: (i) datalog whenever ρ ∈ Σ implies EV(ρ) = /0; (ii) inclusion- dependencies whenever ρ ∈ Σ implies that ρ contains only simple atoms and |body(ρ)| = 1; (iii) linear whenever ρ ∈ Σ implies |body(ρ)| = 1; (iv) guarded whenever ρ ∈ Σ implies that there is an atom of body(ρ) containing all the variables of UV(ρ); (v) joinless whenever ρ ∈ Σ implies that head(ρ) is a simple atom and body(ρ) contains no repeated variables. |
Processed_Finite_model_reasoning_over_existential_rules.txt | position p[i] and Z occurs in the head of ρ at position r[ j]. Ontology Σ belongs to weakly-acyclic if G(Σ) has no cycle going through an arc labeled as special. |
Processed_Finite_model_reasoning_over_existential_rules.txt | Stickiness (Cal`ı et al. 2012). Informally, Σ ∈ sticky guarantees that: if X occurs multiple times in body( ¯ρ), then X occurs in head( ¯ρ) and h(X) belongs to every atom of chase(D, Σ) that de- pends on h(head( ¯ρ)). Formally, a variable X of Σ is marked if (i) there is a rule ρ ∈ Σ such that X occurs in body(ρ) but not in head(ρ); or (ii) there are two rules ρ,ρ′ ∈ Σ such that a marked variable occurs in body(ρ) at some position p[i] and X occurs in head(ρ′) at position p[i] too. Ontology Σ belongs to sticky if, for each ρ ∈ Σ, the following condition is satisfied: if X occurs multiple times in body(ρ), then X is not marked. A more refined condition identifies interesting cases in which it is safe to allow rules containing some marked variable that occurs multiple times but in a single body atom only. This refinement gives rise to sticky-join, generalizing both sticky and linear. |
Processed_Finite_model_reasoning_over_existential_rules.txt | Shyness (Leone et al. 2012). Informally, Σ ∈ shy guarantees that: (1) if X occurs in two dif- ferent atoms of body( ¯ρ), then h(X) ∈ C; and (2) if X and Y occur both in head( ¯ρ) and in two different atoms of body( ¯ρ), then h(X) = h(Y ) implies h(X) ∈ C. Formally, consider an existen- tial variable X of Σ. Position p[i] is invaded by X if there is a rule ρ of Σ such that: (i) X occurs in head(ρ) at position p[i], or (ii) some universal variable Y of ρ is attacked by X —namely Y occurs in body(ρ) only at positions invaded by X— and it also occurs in head(ρ) at position p[i]. A universal variable is protected if it is attacked by no existential variable. Ontology Σ belongs to shy if, for each ρ ∈ Σ, the following conditions are both satisfied: (1) if X occurs in two different atoms of body(ρ), then X is protected; and (2) if X and Y occur both in head(ρ) and in two different atoms of body(ρ), then X and Y are not attacked by the same variable. |
Processed_Finite_model_reasoning_over_existential_rules.txt | In this section we design a general technique to facilitate the process of (dis)proving finite controllability of an arbitrary ontological fragment of existential rules. More specifically, from a triple hD, Σ, qi we build the triple hDc, Σc, qci enjoying the following properties: (1) Dc is propositional database; (2) Σc are constant-free rules containing only simple atoms; (3) qc is a constant-free UBCQ with only simple atoms; (4) chase(Dc, Σc) is a constant-free instance con- taining only simple atoms; and (5) there is a “semantic” correspondence between mods(D, Σ) and mods(Dc, Σc). By exploiting these properties, one can apply the technique shown in Figure 2. |
Processed_Finite_model_reasoning_over_existential_rules.txt | Dc ∪ {p[1](ni)}i>0 ∪ { f[1,c1](n1), f[1,c2](n2)} ∪ { f[1,2](ni+2, ni)}i>0. As a result, the rewriting separates the interaction between the database constants propagated body-to-head via universal variables and the nulls introduced to satisfy existential variables. Also, since the predicates encode the “shapes” of the twin atoms —namely f[1,2](X,Y ) means different nulls while f[1,1](X) the same null— repeated variables are encoded too. By following the same approach, we can rewrite also the query. Consider for example the BCQ q = ∃X p(X), f (X, c1). Therefore, qc is the UBCQ: (p[c1], f[c1,c1]) ∨ (p[c2], f[c2,c1]) ∨ (∃X p[1](X), f[1,c1](X)). |
Processed_Finite_model_reasoning_over_existential_rules.txt | Let us fix a triple hD, Σ, qi through the rest of this section. Consider an atom α = p(t1, . . . ,tm) with terms over C ∪ V. The canonical atom of α is the atom αc = p[ℓ1,...,ℓm](τ1, . . . , τµ), where: (a) ℓi = ti if ti ∈ C; (b) ℓi = ℓ j if ti = t j; or (c) ℓi = 1 + max({0} ∪ {ℓ j ∈ N : j < i}) if ti ∈ V and t j 6= ti ∀ j < i and τi = V ∈ V, if there exists t j such that ℓ j = i and t j = V . Moreover, given a set of atoms A, we define Ac = {αc : α ∈ A}, and give a rule ρ, we define ρc as the rule so that body(ρc) = body(ρ)c and head(ρc) = head(ρ)c. For instance, let α = p(c1, X, c2, X,Y, Z,Y ) be an atom. Then, the canonical atom αc of α is given by p[c1,1,c2,1,2,3,2](X,Y, Z). Note that, by definition of τi, for i = 1, . . . ,µ, we have that the arity µ ≤ m of the canonical atom is equal to max({0} ∪ { f (t j) ∈ N : j ≤ m}). |
Processed_Finite_model_reasoning_over_existential_rules.txt | Definition 3.1 (Safe and Canonical substitutions) A map ς : const(D ∪ Σ) ∪ V → const(D ∪ Σ) ∪ V is called canonical substitution if ς(c) = c for each c ∈ const(D ∪ Σ). Moreover, we say that a canonical substitution ς is safe w.r.t. a rule ρ ∈ Σ if ς(UV(ρ)) ⊆ const(D ∪ Σ) ∪ UV(ρ), and ς(V ) = V , for each V ∈ EV(ρ). |
Processed_Finite_model_reasoning_over_existential_rules.txt | Intuitively, a safe substitution maps each existential variable to itself and no universal vari- able is mapped to an existental one. As usual, given a set of atoms A, we denote by ς(A) = {p(ς(t1), . . . ,ς(tm)) : p(t1, . . . ,tm)}, and given a rule ρ, we denote by ς(ρ) the rule such that body(ς(ρ)) = ς(body(ρ)) and head(ς(ρ)) = ς(head(ρ)). |
Processed_Finite_model_reasoning_over_existential_rules.txt | We denote by CS the set of all canonical substitutions and by ß(ρ) ⊆ CS the set of all safe substitutions w.r.t. ρ. Given a set of atoms A [resp. a rule ρ] and a canonical substitu- tion [resp. safe substitution] ς, we say that ς(A)c [resp. ς(ρ)c] is the canonical set of atoms w.r.t A [resp. canonical rule w.r.t. ρ] and ς. Observe that two different canonical substitutions could produce two isomorphic canonical set of atoms. For instance, let A = {p(X,Y )}, and con- sider ς = {X 7→ X, Y 7→ Y } and ς′ = {X 7→ Y, Y 7→ X}. Then, ς(A)c = {p[1,2](X,Y )}, and ς′(A)c = {p[1,2](Y, X)} are isomorphic set of atoms. Therefore, to avoid redundancies, we denote by CS∗ [resp. ß∗(ρ)] any arbitrary maximal subset of CS [resp. of ß(ρ)] producing canonical set of atoms [resp. canonical rules] containing no two isomorphic elements. |
Processed_Finite_model_reasoning_over_existential_rules.txt | We denote by Σc the set of all canonical rules {ς(ρ)c : ρ ∈ Σ and ς ∈ ß∗(ρ)}, and we call it the canonical rewriting of Σ. Also, given a UBCQ q of the form ∃Y1ψ1(Y1) ∨ . . . ∨ ∃Ykψk(Yk), we denote by qc the disjunction Wς1∈CS∗ ς1(ψ1(Y1))c ∨ . . . ∨ Wςk∈CS∗ ςk(ψk(Yk))c. and we call it the canonical rewriting of q. Finally, we call Dc the canonical rewriting of D. |
Processed_Finite_model_reasoning_over_existential_rules.txt | Proposition 3.1 The triple hDc, Σc, qci can be constructed from hD, Σ, qi in polynomial time (in data complexity). |
Processed_Finite_model_reasoning_over_existential_rules.txt | Proposition 3.2 Consider a set Σ of existential rules. For each database D and for each UBCQ q, it holds that R(chase(Dc, Σc)) = chase(D, Σ) and R(qc) ≡ q. |
Processed_Finite_model_reasoning_over_existential_rules.txt | By exploiting the above proposition, we can now prove that a UBCQ q is satisfied by all models of a theory D ∪ Σ if, and only if, each model of the canonical rewriting of the theory Dc ∪ Σc satisfies the canonical rewriting of the UBCQ qc. |
Processed_Finite_model_reasoning_over_existential_rules.txt | Theorem 3.1 D ∪ Σ |= q if, and only if, Dc ∪ Σc |= qc. |
Processed_Finite_model_reasoning_over_existential_rules.txt | Note that, if Σ is a constant-free ontology, then, for each model Mc of Dc ∪ Σc, R(Mc) is a model of D ∪ Σ. The request for a constant-free ontology is needed. Indeed, for instance, let Σ = {p(a) → r(a); r(x) → p(x)}. So that, Σc = {p[a] → r[a]; r[a] → p[a]; r[1](V1) → p[1](V1)}. Therefore, Mc = {p[1](a)} is a model of Σc, but R(Mc) = {p(a)} is not a model of Σ, as it does not satisfy the first rule. However, we can overcome this problem considering the following class of models. Given a model Mc ∈ mods(Dc, Σc), we say that Nc is a smooth instance of Mc if there exists a bijective map f : terms(Mc) → terms(Nc) such that f (n) = n for each null n ∈ terms(Mc); f (c) = nc for each constant c ∈ terms(Mc), where nc is a fresh null; and f (Mc) = Nc. Note that a smooth instance of a model Mc is also a model of Dc ∪ Σc and it is also constant-free. |
Processed_Finite_model_reasoning_over_existential_rules.txt | Proposition 3.3 If Mc ∈ mods(Dc, Σc), then R(Nc) ∈ mods(D, Σ), for each smooth model Nc of Mc. |
Processed_Finite_model_reasoning_over_existential_rules.txt | By exploiting the above proposition, we can now prove that a UBCQ q is satisfied by all finite models of a theory D ∪ Σ if, and only if, each finite model of the canonical rewriting of the theory Dc ∪ Σc satisfies the canonical rewriting of the UBCQ qc. |
Processed_Finite_model_reasoning_over_existential_rules.txt | Theorem 3.2 D ∪ Σ |=fin q if, and only if, Dc ∪ Σc |=fin qc. |
Processed_Finite_model_reasoning_over_existential_rules.txt | Mc ∈ fmods(Dc, Σc) such that R(Mc) = M. Hence, let h be a homomorphism from qc to Mc. So that, h(ςj(ψj(Y j))c) ⊆ Mc, for some disjunct ςj(ψj(Y j))c of qc. Therefore, by applying the unpacked function, we have that R(h(ςj(ψj(Y j))c)) = h(R(ςj(ψj(Y j))c)) = h(ςj(ψj(Y j)) ⊆ R(Mc) = M. Hence, h is also a homomorphism from q to M. |
Processed_Finite_model_reasoning_over_existential_rules.txt | By exploiting the properties of the canonical rewriting, one can reprove that linear is finitely controllable, and prove (for the first time) that also sticky-join enjoys this property. In fact, given a linear or sticky-join ontology Σ, its canonical rewriting Σc belongs to inclusion-dependencies or sticky, respectively. In the former case, it suffices to observe that any variable occurring multiple times in some atom α, by definition, occurs exactly once in its associated canonical atom αc. In the latter case, additionally, consider a variable X violating the sticky property since it is marked and it occurs multiple times in the body of some rule ρ. By hypothesis, X may occur in exactly one atom of body(ρ). However, even if marked, X now occurs exactly once in its canonical atom and it cannot violate the sticky property any more. The following result follows. |
Processed_Finite_model_reasoning_over_existential_rules.txt | Theorem 3.3 Under sticky-join ontologies, D ∪ Σ |= q if, and only if, D ∪ Σ |=fin q. |
Processed_Finite_model_reasoning_over_existential_rules.txt | With these tools in place, we can now apply the technique shown in Figure 3, where we use the symbol |=wsf to refer the satisfiability of the query under well-supported finite models only. |
Processed_Finite_model_reasoning_over_existential_rules.txt | As said, and next stated, the canonical rewriting of a shy ontology is again a shy ontology. |
Processed_Finite_model_reasoning_over_existential_rules.txt | We start by defining the notion of well-supported finite instances, which is inspired by the related notion of well-supported interpretations for general logic programs (Fages 1991). |
Processed_Finite_model_reasoning_over_existential_rules.txt | Let D be a database, and Σ be an ontology. A finite instance I is called well-supported w.r.t. the theory D ∪ Σ if there is an ordering (α1, . . . ,αm) of its atoms such that, for each j ∈ {1, . . . , m}, at least one of the following conditions is satisfied: (1) αj is a database atom of D; and (2) there exist a rule ρ of Σ and a homomorphism h from atoms(ρ) to {α1, . . . ,αj} such that h(head(ρ)) = {αj} and h(body(ρ)) ⊆ {α1, . . . ,αj−1}. In both cases, we will say that αj is a well-supported atom w.r.t. (α1, . . . ,αm); while in the latter case we will also say that ρ is a well-supporting rule for αj w.r.t. (α1, . . . ,αm). Such an ordering will be called a well-supported ordering of I. |
Processed_Finite_model_reasoning_over_existential_rules.txt | We denote by wsfmods(D, Σ) ⊆ fmods(D, Σ) the set of all well-supported finite models of D ∪ Σ. Moreover, if a UBCQ q is satisfied by each model of wsfmods(D, Σ), we write D ∪ Σ |=wsf q. Interestingly, each finite model of D ∪ Σ contains a well-supported finite model of the theory. |
Processed_Finite_model_reasoning_over_existential_rules.txt | Proposition 4.2 For each M ∈ fmods(D, Σ), there exists M′ ⊆ M such that M′ ∈ wsfmods(D, Σ). In particular, each minimal finite model of D ∪ Σ is a well-supported finite model. |
Processed_Finite_model_reasoning_over_existential_rules.txt | Although each finite model of an ontological theory contains a well-supported finite model of the theory, the reverse inclusion does not hold. Consider for example the ontology Σ of Sec- tion 3.1, and the model M = D ∪ { f (c1, c1), f (c2, c1)}. Since (p(c1), p(c2), f (c1, c2), f (c1, c1), f (c2, c1)) is a well-supported ordering of M, then M is well-supported. However, M \ { f (c2, c1)} is a model of D ∪ Σ. Therefore, M is not a minimal one. Using Proposition 4.2, we can now prove that if a UBCQ q can be satisfied by each well-supported finite model of a theory, then it can be satisfied by each finite model of the theory. |
Processed_Finite_model_reasoning_over_existential_rules.txt | Theorem 4.2 D ∪ Σ |=wsf q if, and only if, D ∪ Σ |=fin q. |
Processed_Finite_model_reasoning_over_existential_rules.txt | Proof Clearly, by subset inclusion, if each finite model of D ∪ Σ satisfies the query q, then each well- supported finite model of D ∪ Σ satisfies q. Moreover, as each finite minimal model is a well- supported finite model (Proposition 4.2), then for each finite model M′ of D ∪ Σ, we can find a well-supported finite model, that is minimal, M of D ∪ Σ, such that M ⊆ M′, and, in particular, there exists a homomorphism h (i.e., the identity homomorphism) such that h(M) ⊆ M′. |
Processed_Finite_model_reasoning_over_existential_rules.txt | Let us start with the preliminary notions of existentially well-supported atom and propagated term. Let I be a well-supported finite instance, and (α1, . . . ,αm) be a well-supported ordering of I. An atom α of I \ D is said existentially well-supported w.r.t. the ordering (α1, . . . ,αm) if, for each well-supporting rule ρ for α w.r.t. (α1, . . . ,αm), it holds that EV(ρ) 6= /0. Moreover, let αj[k] = t, for some position k, then t is said propagated from an atom αi in position l, whenever i < j, αi[l] = t, and there exist a well-supporting rule ρ for αj and a homomorphism h such that αi ∈ h(body(ρ)). Consider again ontology Σ of Section 3.1, and the well-supported finite model M considered after Proposition 4.2. For instance, the atom f (c1, c1) is existentially well- supported. Indeed, the unique way to well-support the atom comes from the first rule of Σ, that is an existential rule. We are now ready to define the notion of propagation ordering. |
Processed_Finite_model_reasoning_over_existential_rules.txt | Definition 4.1 (Propagation ordering) Let D be a database, Σ be a joinless ontology, M ∈ wsfmods(D, Σ), and (α1, . . . ,αm) be a well- supported ordering of M. For each αj ∈ M, we build a new atom hαji as follows. Let t = αj[k]. We have: (1) If αj is an existentially well-supported atom and k is an existential position, then hαji[k] = ht, j, ki, where ht, j, ki is called a starting point of t; (2) If t is a propagated term from some atom αi in position l, then hαji[k] = hαii[l]; and (3) hαji[k] = αj[k], otherwise. We call (hα1i, . . . , hαmi) a propagation ordering of the well-supported ordering (α1, . . . ,αm). |
Processed_Finite_model_reasoning_over_existential_rules.txt | useful to remember a starting point of that particular term and its propagations in other atoms. |
Processed_Finite_model_reasoning_over_existential_rules.txt | With our technical tools in place, we are now able to prove the following technical result. |
Processed_Finite_model_reasoning_over_existential_rules.txt | Proof intuition Consider an arbitrary model M ∈ wsfmods(Dc, Σc a). It suffices to prove that there exist M′ ∈ wsfmods(Dc, Σc) and a homomorphism h′ s.t. h′(M′) ⊆ M. Indeed, by hypothesis, there exists a homomorphism h s.t. h(q) ⊆ M′, and so (h′ ◦ h)(q) ⊆ M. |
Processed_Finite_model_reasoning_over_existential_rules.txt | One can verify that M = {s[c], p[1](n1), r[1](n1)} is a (minimal) well-supported finite model of Dc ∪ Σc a is obtained from Σc by discarding the last harmless rule. However, M is not a model of Dc ∪ Σc because the last rule is not satisfied. |
Processed_Finite_model_reasoning_over_existential_rules.txt | The idea is to show how to construct from M a model M′ ∈ wsfmods(Dc, Σc) that can be homomorphically mapped to M. Intuitively, we identify the starting points in which existential variables of Σc a have been satisfied and rename the introduced terms using a propagation ordering. In the example above, consider the well-supported ordering (s[c], p[1](n1), r[1](n1)) of M, re- place n1 in p[1](n1) by hn1, 2, 2i (null n1 introduced in the second atom in the second position), and replace n1 in r[1](n1) by hn1, 3, 2i (null n1 introduced in the third atom in the second posi- tion). Then, since M is well-supported, we propagate (if needed) these new terms according the supporting ordering. In our case, M′ = {s[c], p[1](hn1, 2, 2i), r[1](hn1, 3, 2i)} is now a finite model of Dc ∪ Σc that can be mapped to M. |
Processed_Finite_model_reasoning_over_existential_rules.txt | Lemma 4.1 Under shy ontologies, Dc ∪ Σc |= qc if, and only if, Dc ∪ Σc |=fin qc. |
Processed_Finite_model_reasoning_over_existential_rules.txt | Proof Clearly, the “only if” implication is straightforward. Hence, given a shy ontology Σ, we have to prove that Dc ∪ Σc |= qc, whenever Dc ∪ Σc |=fin qc, for each database D and UBCQ q. Suppose that Dc ∪ Σc |=fin qc, i.e., the query qc is satisfied by each finite model of Dc ∪ Σc. Thus, by Theorem 4.2, holds that Dc ∪ Σc |=wsf qc, that is, the canonical rewriting of the query q is satisfied by each well-supported finite model of the logical theory Dc ∪ Σc. Then, by Theorem 4.3, holds a |=wsf qc, that is, the canonical rewriting of the query q is satisfied by each well- that Dc ∪ Σc supported finite model of the joinless logical theory Dc ∪ Σc a. Moreover, again, by Theorem 4.2, a |=fin qc, that is qc is satisfied also by every finite model of the previous we obtain that Dc ∪ Σc theory. Now, as Σc a is a joinless ontology, by the finite controllability of joinless ontologies proved a |= qc. Finally, by Theorem 4.1, we have by Gogacz and Marcinkowski (2013), holds that Dc ∪ Σc that Dc ∪ Σc |= qc, i.e. the query qc is satisfied by each model (finite or infinite) of Dc ∪ Σc. |
Processed_Finite_model_reasoning_over_existential_rules.txt | To complete the related works started with the Introduction, we recall that the notion of finite controllability was formalized for the first time by Rosati (2006) while he was working on a question that had been left open two decades before by Johnson and Klug (1984) about contain- ment of conjunctive queries in case of both arbitrary and finite databases. Basically, using our terminology, they proved that ontologies mixing both inclusion-dependencies and functional- dependencies are not finitely controllable, by leaving open the case where ontologies contain inclusion-dependencies only. Rosati then answered positively this question. |
Processed_Finite_model_reasoning_over_existential_rules.txt | The semantic equivalence of fundamental reasoning tasks under finite and infinite models is not at all a prerogative of the database community. A sister yet orthogonal property of finite controllability is of paramount importance also in logic, where it has been investigated much earlier. It is known as finite model property or finite satisfiability (Ebbinghaus and Flum 1995), and it asks for a class C of sentences whether every satisfiable sentence of C has a finite model. For example, both G¨odel and Sch¨utte proved that ∀2∃∗ first-order sentences are finitely satisfiable. |
Processed_Finite_model_reasoning_over_existential_rules.txt | Although reasoning under finite models has a long history and it has been actively investi- gated in various fields of Computer Science, finite controllability remains open for many lan- guages combining or generalizing the key properties underlying the basic classes depicted in Figure 1. By way of example, we mention (i) glut-guarded (Kr¨otzsch and Rudolph 2011), ex- tending guarded and weakly-acyclic; (ii) weakly-sticky-join (Cal`ı et al. 2012), extending sticky- join, weakly-acyclic and shy; and (iii) tame (Gottlob et al. 2013), extending sticky and guarded. Between guarded and glut-guarded, it is worth to recall weakly-guarded (Cal`ı et al. 2013), where each rule body has an atom covering all those variables that only occur in invaded (a.k.a. affected) positions. Actually, this class is finitely controllable although the proof sketch given by B´ar´any et al. (2014) has some hole (there, some model of D ∪ Σ′ might not satisfy Σ). In fact, our canonical rewriting yields an ontology that can be partitioned in active and harmless, where the active part is guarded. Well-supported models and propagation ordering behave as for shy. |
Processed_Finite_model_reasoning_over_existential_rules.txt | since they are not standard (actually stricter). In the former, repeated variables are admitted only in rule heads, while for the latter the authors state that the difference between sticky and sticky- join “can only be seen if repeated variables in the heads of the rules are allowed”. (Regarding sticky, the classical notion is only rephrased: their “immortal” positions correspond to positions being not marked.) From such a mismatch, however, it follows that finite controllability of sticky- join was unknown before our work. A curious reader may verify that the proof of their Lemma 4 breaks down when moving to a linear (hence sticky-join) ontology such as Σ = {p(X, X) → r(X); r(X) → ∃Y r(Y )} —inducing no immortal position since all positions p[1], p[2] and r[1] host marked variables— paired with the singleton database D = {p(c, c)}. |
Processed_Finite_model_reasoning_over_existential_rules.txt | By demonstrating that shy is finitely controllable, we complete an important picture around the basic decidable Datalog± classes. But we take it as a starting point rather than an ending one. |
Processed_Finite_model_reasoning_over_existential_rules.txt | On the one hand, finite controllability immediately implies decidability of OBQA. Actually, via the soundness and completeness of the chase procedure we know that the problem of deciding whether a UBCQ is true over a Datalog± theory is recursively enumerable. But the complemen- tary problem of deciding whether a UBCQ is false over a finitely controllable Datalog± class C is recursively enumerable too. In fact, each theory D ∪ Σ, with Σ ∈ C, always admits a fair lexicographic enumeration of its finite models. Unfortunately, such a na¨ıve procedure would be inefficient in practice. Making it usable and competitive for real world problems is challenging and it is part of our ongoing work. Basically, this would lead to a tool able to deal with any finitely controllable fragment, some of which (e.g., guarded) have no effective implementation. On the other hand, we believe the techniques developed in this paper could have future ap- plications. For example, we are working on an extended version of our canonical rewriting that encodes in the predicates also a limited amount of nulls. This requires more complex techniques, which however would apply to classes using the key properties underlying weakly-acyclic, such as glut-guarded and weakly-sticky-join (see Section 5). Hence, by combining these techniques with the above tool for finitely controllable classes, we aim at the design and implementation of a reasoner able to deal with ontologies falling in any known decidable Datalog± class. |
Processed_Finite_model_reasoning_over_existential_rules.txt | Finally —even if the unrestricted set of existential rules cannot be finitely controllable since it is not decidable— it is still open, to the best of our knowledge, whether there exists, or not, a fragment of existential rules which is decidable but not finitely controllable. |
Processed_Finite_model_reasoning_over_existential_rules.txt | The paper has been partially supported by the Italian Ministry for Economic Development (MISE) under project “PIUCultura – Paradigmi Innovativi per l’Utilizzo della Cultura” (n. F/020016/01- 02/X27), and under project “Smarter Solutions in the Big Data World (S2BDW)” (n. F/050389/01- 03/X32) funded within the call “HORIZON2020” PON I&C 2014-2020. |
Processed_Finite_model_reasoning_over_existential_rules.txt | GOGACZ, T. AND MARCINKOWSKI, J. 2017. Converging to the chase - A tool for finite controllability. |
Processed_Finite_model_reasoning_over_existential_rules.txt | Let φ(X) be a conjunction of atoms, and let X ∈ X. We say that X is attacked by a variable Y in φ(X) if all the positions where X appears are invaded by Y . On the other hand, we say that X is protected in φ(X), if it is attacked by no variable. |
Processed_Finite_model_reasoning_over_existential_rules.txt | Finally, if each ρ ∈ Σ is shy w.r.t. Σ, then call Σ a shy ontology. |
Processed_Finite_model_reasoning_over_existential_rules.txt | Let Σ = {ρ1,ρ2,ρ3}. Clearly, ρ1 and ρ3 are shy rules w.r.t. Σ, since they are also linear rules, namely rules with one single body atom, which cannot violate any of the two shy conditions. Moreover, rule ρ2 is also shy w.r.t. Σ as the positions p[2] and u[1] are invaded by disjoint sets of existential variables. Indeed, p[2] is invaded by the existential variable Y1 of the first rule, and u[1] is invaded by the existential variable Y3 of the third rule. Therefore, Σ is a shy ontology. |
Processed_Finite_model_reasoning_over_existential_rules.txt | Let Σ′ be the ontology Σ ∪ {ρ4}. It is easy to see that ρ1, ρ3 and ρ4 are shy w.r.t. Σ′. However, ρ2 is not shy w.r.t. Σ′, as property (i) is not satisfied. Indeed, the variable Y2 occurring in two body atoms in body(ρ2) is not protected, as the position p[2] and u[1] (the only positions in which Y2 occurs) are invaded by the same existential variable, namely Y3. Therefore, Σ′ is not a shy ontology. |
Processed_Finite_model_reasoning_over_existential_rules.txt | Both the new rules satisfy now condition (i) w.r.t. Σ′. Moreover, it is not difficult to see that, for every database D and for every UBCQ q, it holds that D ∪ Σ |= q if and only if D ∪ Σ′ |= q. However, since ρ does not satisfy condition (i), this immediately implies that the first new rule does not satisfy condition (ii). |
Processed_Finite_model_reasoning_over_existential_rules.txt | The syntactic properties of shy make the class quite expressive since it strictly contains both linear and datalog. Moreover, these properties are easy recognizable and guarantee efficient an- swering to conjunctive queries, as experimentally shown in Leone et al. (2012). In fact, ontology- based query answering over shy ontologies preservers the same data and combined complexity of OBQA over datalog, namely PTIME-complete and EXPTIME-complete, respectively. |
Processed_Finite_model_reasoning_over_existential_rules.txt | Proof of Proposition 3.2 We prove that R(chase(Dc, Σc)) = chase(D, Σ) by induction on the chase step. Let I0 = D ⊂ I1 ⊂ . . . ⊂ Im ⊂ . . . be a chase procedure of D and Σ; and let Ic m ⊂ . . . be a chase procedure of Dc and Σc. |
Processed_Finite_model_reasoning_over_existential_rules.txt | Finally, let qc be the canonical rewriting of the UBCQ q = ∃Y1ψ1(Y1) ∨ . . . ∨ ∃Ykψk(Yk). For each j ∈ {1, . . . , k}, consider the safe substitution ςj mapping each variable of ψj(Y j) in a dif- j (Y j) = ςj(ψj(Y j))c in qc, ferent null. Therefore, there exists a conjunction of atoms, say ψc such that R(ψc j (Y j)) = ψj(Y j), for each j ∈ {1, . . . , k}. Hence, q ⊆ R(qc). Moreover, it is easy to see that, each other safe substitution ς′ w.r.t. some ψj, produces a conjunction of atoms, ς′(ψj(Y j))c such that R(ς′(ψj(Y j))c) is contained in R(ςj(ψj(Y j))c). Therefore, R(qc) ⊆ q. Thus, R(qc) = q. |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.