![List of Figures
1.1 The Kompaı̈-2 robot and its user interface. 10
2.1 The Semantic Web stack. 17
2.2 An example of collected uses-story. 24
3.1 The eXtreme Design workflow as extended in [164]. The
highlighted tasks involve the guidelines for ontology
reuse. 33
3.2 The template of the user stories provided by the cus-
tomer representatives. 38
3.3 The network of ontologies constituting the MARIO
Ontology Network. 39
3.4 The diagram of the Affordance ontology. 46
3.5 Two equivalent action-selection schemes. 49
3.6 The UML class diagram of CGA ontology. 53
3.7 The UML class diagram of the Co-Habitation status
ontology. 55
3.8 The UML class diagram of the Medication Use ontology. 56
3.9 The UML class diagram of the Capability Assessment
ontology. 57
3.10 The UML class diagram of the SPMSQ ontology. 59
3.11 The UML class diagram of the ESS ontology. 60
3.12 The UML class diagram of the CIRS ontology. 62
3.13 The UML class diagram of the MNA ontology. 62
3.14 The UML class diagram of tagging ontology. 64
3.15 The UML class diagram of Music ontology. 70
4.1 An example of Framester representation for the concept
G suit. 80
xi](https://image.slidesharecdn.com/18607690-250603002415-ccda2ab2/85/Engineering-Background-Knowledge-For-Social-Robots-Luigi-Asprino-17-320.jpg)
![Chapter 1
Introduction
Social robots [12, 27, 46, 54, 51, 67] are autonomous embodied agents
that interact, collaborate, communicate with humans, by following
the behavioral norms expected by people with whom robots are in-
tended to interact. Several definitions have been proposed for the
term “social robot”, but all of them broadly agree that a social ro-
bot has the following characteristics: (i) Physical embodiment, i.e. a
social robot has a physical body; (ii) Sociality, i.e. a social robot is
able to interact with people by showing human-like features while
following the social rules (defined through society) attached to its
role; (iii) Autonomy, i.e. a social robot makes decisions by itself (the
autonomy is sometimes limited in testing phase, like in the Wizard
of Oz experimental setting [100, 168]). In recent years, the field of
socially assistive robotics [62] has emerged given the great potential
of social robots in supporting people with cognitive impairment or
physical disability [17, 86, 94, 139]. As overviewed in [127], existing
robotic technologies for care range from pet-like devices to advanced
anthropomorphic mobile robotic assistants. While service robots often
focus on providing physical support, a socially assistive robot aims to
provide cognitive support through social interaction.
In order to effectively interact, communicate, collaborate with
humans, robots should demonstrate some intelligence. Generally, the
approaches for building such an intelligence can be classified into two
categories, namely, symbolic and subsymbolic approaches [159, 185].
In symbolic approaches real-world entities, facts, rules and concepts
are formalized by means of symbols. Symbols are stored in a knowledge
base and are manipulated to make conclusions and take decisions.
1](https://image.slidesharecdn.com/18607690-250603002415-ccda2ab2/85/Engineering-Background-Knowledge-For-Social-Robots-Luigi-Asprino-21-320.jpg)
![2 CHAPTER 1. INTRODUCTION
Conversely, subsymbolic approaches attempt to address problems
without building an explicit representation of concepts and entities. An
example of subsymbolic approaches are artificial neural networks, i.e.
large networks of extremely simple numerical processors (i.e. neurons),
massively interconnected and running in parallel. These networks
consume and produce numerical vectors. The connections between
neurons determine how input vectors is transformed to an output
vector. In these systems, knowledge is not encoded with symbols
but rather in the pattern of numerical strengths of the connections
between neurons. These networks can “learn” to perform a given task
by seeing a set of examples. An example is a pair of a possible input for
the task at hand with (possibly) the corresponding desirable output.
From these examples a neural network can learn the optimal weights
to predict the desirable output from a given input. There are benefits
and drawbacks of both kinds of approaches. While subsymbolic
approaches achieve better performance in specific tasks, they need a
large quantity of examples to learn optimal strengths. These examples
could be hardly available for certain tasks. The other main issue
with subsymbolic systems is the inability of explaining why a certain
output is provided for a given input. This could be a non-negligible
problem for many domain, such as medicine. In contrast, symbolic
systems are able to explain their decisions and do not require examples.
However, these systems need considerable effort for defining symbols
and designing rules and methods to manipulate symbols for solving
problems.
Integrating the two approaches have also become more common
to benefit of both strategies (e.g. [82, 89, 134]). The framework
proposed in this thesis benefits of both symbolic and subsymbolic
techniques. Subsymbolic techniques are used in perceptual tasks
(such as translation of spoken language into text), whereas symbolic
techniques are used for controlling the robot at an higher level [83]. In
particular, subsymbolic subsystems of the robot transform low-level
perception in symbols so to enable the symbolic processing of the
control system. In this way the framework benefits of the state-of-the-
art performance on perceptual tasks of subsymbolic techniques without
compromising the possibility of having a system that is deterministic
and able to explain its behavior and decisions (important requirements
for the case study of this thesis, cf. Section 1.3).
In order to show human-like feature the robot should be able to](https://image.slidesharecdn.com/18607690-250603002415-ccda2ab2/85/Engineering-Background-Knowledge-For-Social-Robots-Luigi-Asprino-22-320.jpg)
![4 CHAPTER 1. INTRODUCTION
architectures. The choice of a knowledge representation formalism has
an impact on the expressiveness (the breadth of concepts that can be
represented), on the ability infer logical consequences form asserted
facts in a tractable manner (i.e. tractability) and on the amount of
tools and knowledge already available that can be provided to robots
for supporting their tasks.
Predicate logic is a collection of formal systems that includes: pro-
positional logic (dealing with zero-arity predicates, called propositions),
first-order logic (FOL) (dealing with terms of higher arity) and higher
order logic (in which predicates can be arguments of other predicates).
Unlike propositional logic, first-order logic and higher order logic are
undecidable (i.e. it does not exist a method for validating all the
formulas). To overcome the undecidability of FOL, some fragments of
the logic have been proposed (e.g. horn clause and description logic).
Propositional logic, horn clause (which constitutes the foundation
of logic programming languages), description logic [8] (at the basis
of the W3C’s OWL-DL standard) are used in several robotic frame-
works (e.g. [131, 142, 176, 196]). Modal logic extends propositional
and predicate logic to include the modality operator. A modal is an
operator that qualifies statements (e.g. usually, possibly). An example
of robotic framework based on modal logic is Tino [50]. Temporal
logic allows to qualify proposition in terms of time (e.g. “I am always
hungry”). It is used in some robotic frameworks such as [59, 104].
Another logics particularly relevant for robotic frameworks (e.g. [111])
is the probabilistic logic which provides a formalism able to handle
uncertainty.
All of these logic frameworks allow to potentially encode robot’s
knowledge and are supported by several off-the-shelf reasoning tools.
Although some of them show high expressive power and attractive
computational features, they almost lack of existing resources that
could provide robots with a comprehensive background knowledge for
their tasks. The choice of a framework impacts on the effort required
to provide a robot with a considerable amount of knowledge for its
tasks. This problem is even more significant for social robots which
need of a sizable and heterogeneous knowledge bases to operate.
The Semantic Web [18] standards give a good trade-off among the
expressiveness, tractability and availability of resources and tools for
manipulating such knowledge. The Semantic Web (cf. Section 2.2) is
an extension of the World Wide Web aimed at providing a framework](https://image.slidesharecdn.com/18607690-250603002415-ccda2ab2/85/Engineering-Background-Knowledge-For-Social-Robots-Luigi-Asprino-24-320.jpg)
![1.1. GOALS OF THE THESIS 5
that allows data to be shared with a common syntax and semantics.
It based on a language defining the syntax of data to be shared (i.e.
XML), a model defining the format of data (i.e. RDF) and a language
to formally specify the semantics of data (i.e. OWL). OWL is derived
from description logics and comes in three forms: (i) OWL Full for
maximum expressiveness, but undecidable; (ii) OWL DL designed
to provide the maximum expressiveness while retaining decidability;
(iii) OWL Lite that supports taxonomies and simpler constraints with
a limited expressiveness but attractive computational features. Se-
mantic Web standards provide a good expressive power (equivalent to
Description Logics), without compromising decidability and tractabil-
ity. It is supported by a pletora of off-the-shelf reasoners, knowledge
management systems, as well as tools for creating, organizing and
integrating knowledge. Another benefit of implementing a framework
that relies on Semantic Web technologies is the opportunity of ex-
ploiting knowledge available as Linked Open Data (LOD). LOD is a
huge web of data (∼200 billion linked facts1
) formally (and uniformly)
represented in RDF and OWL, and openly available on the Web.
1.1. Goals of the Thesis
The aim of this thesis can be summarized in the following question.
RQ0: To what extent Semantic Web technologies and Linked Data can
be used to create, organize, access to, and evolve robot’s background
knowledge?
Existing solutions for managing robots’ knowledge (such as RoboE-
arth [198], RoboBrain [176], ORO [113], ORA [161], RACE [172],
and OUR-K framework [117]) only partially exploit the potential
of Semantic Web technologies and Linked Data. In these projects,
Semantic Web technologies are mostly employed to address syntactic
heterogeneity of data, to define conceptualizations of the robots’
knowledge, and, more rarely, to include external datasets within the
robots’ knowledge base. Most of these frameworks focused on a single
dimension of knowledge while disregarding more social aspects. In fact
knowledge manipulated by these frameworks are often related to the
interaction of the robot with the physical environment for supporting
navigation [142], inter-robot communication [198], manipulation of
1
LODCloud, https://lod-cloud.net/](https://image.slidesharecdn.com/18607690-250603002415-ccda2ab2/85/Engineering-Background-Knowledge-For-Social-Robots-Luigi-Asprino-25-320.jpg)
![6 CHAPTER 1. INTRODUCTION
the objects [196], representation of robot’s actions [199].
We claim that robots’ architectures can profoundly benefit of Se-
mantic Web technologies and Linked Data paradigm. (i) Semantic
Web technologies could enable an incremental and iterative develop-
ment of the architecture. (ii) Semantic Web standards allow to easily
integrate data generated by a variety of components, thus enabling
robots to make decisions by taking into account knowledge about the
physical world, data coming from their operating environment, inform-
ation about social norms, users’ preferences and so on. (iii) Semantic
Web technologies provide flexible solutions to extend and evolve robots’
knowledge over time. (iv) Linked (Open) Data paradigm lets to easily
reuse (i.e. integrate with robots’ knowledge) existing external datasets
so to bootstrap knowledge base with relevant information for robots’
activities. (v) Linked Data also provides a mechanism that allows
robots to mutually share knowledge.
As anticipated with the main research question RQ0, the goal
of this thesis is to investigate feasibility and benefits of engineering
background knowledge of social robots with a framework based on
Semantic Web technologies and Linked Data. This investigation has
been carried out in a socially assistive context (presented in Section 1.3)
by following the methodology described in Section 1.4. The research
question RQ0 can be decomposed into the following sub-questions
which will be investigated and discussed in the following chapters (one
question for each Chapter, from Chapter 3 to Chapter 7).
RQ1: What kind of knowledge a robot needs to operate in socially
assistive context? What exiting ontologies can be used to or-
ganize the robot’s knowledge? What ontologies need to be
advanced? What domains of interest in this context miss of a
conceptualization?
RQ2: What Linked Data can provide background knowledge for social
robots tasks?
RQ3: How to provide robots with access to knowledge?
RQ4: How to integrate robot’s knowledge with data coming from
robot’s experience?
RQ5: How Semantic Web technologies can be orchestrated to support
robot tasks?](https://image.slidesharecdn.com/18607690-250603002415-ccda2ab2/85/Engineering-Background-Knowledge-For-Social-Robots-Luigi-Asprino-26-320.jpg)
![1.2. CONTRIBUTIONS OF THE THESIS 7
1.2. Contributions of the Thesis
This thesis contributes to goals presented in Section 1.1 as follows.
• We have devised a set of interconnected and modularised onto-
logies, i.e. the MARIO Ontology Network (MON), which are
meant to model all knowledge areas that are relevant for robots’
activities in socially assistive contexts (cf. RQ1). This ontology
network defines reference models for representing and structur-
ing the knowledge processed by the robot. MON provides a
robot with the means for creating, organizing, querying and
reasoning over a background knowledge. MON reuses and integ-
rates state-of-the-art ontologies in various domains (such those
related to personal information, social and multimedia contents),
and, proposes novel solutions in medical domain (e.g. CGA
Ontology, cf. Section 3.3.2) and in robotic domain (i.e. Afford-
ance Ontology, cf. Section 3.3.1). These modules enables robots
to assess the medical, psycho-social and functional status of a
person and to decide the most appropriate action to perform in
a given situation.
• The knowledge base is originally populated with lexical, linguistic
and factual knowledge retrieved from Linked Open Data. The
thesis presents a novel process for generating, integrating and
assessing this knowledge (cf. RQ2). Moreover, we propose a
novel empirical method for assessing foundational distinctions
over Linked Open Data entities from a common sense perspective
(e.g. deciding if an entity inherently represents a class or an
instance from a common sense perspective). This method realizes
the first step of a more general procedure meant to automatically
generate common sense knowledge from Linked Open Data (cf.
RQ2). These methods advance state-of-the-art in Semantic Web
by proposing standardized techniques for creating an integrated
repository of linguistic, factual, encyclopedic, ontological and
common sense knowledge. The benefits of these techniques as
well as the resulting datasets are not limited to robotic domain,
but also extend to every application domain that requires a rich
knowledge base to operate [49].
• We have developed an object-RDF mapper (called Lizard) that
facilitates software components to interact with an RDF know-](https://image.slidesharecdn.com/18607690-250603002415-ccda2ab2/85/Engineering-Background-Knowledge-For-Social-Robots-Luigi-Asprino-27-320.jpg)
![8 CHAPTER 1. INTRODUCTION
ledge base (cf. RQ3). In particular, given an RDF knowledge
base and an OWL ontology describing its structure, Lizard
provides applications with an API for accessing RDF facts stored
in a knowledge base following the object-oriented paradigm.
The API reflects the semantics of the input ontology and al-
lows transparent access to the knowledge base, Differently from
existing systems, Lizard exposes the API following the REST
architectural style over HTTP. This tool is aimed at easing the
software development of knowledge-aware systems by filling the
gap between Semantic Web technologies and Object-Oriented
applications. The benefits of Lizard are not to be intended only
for robotic domain, but every application that need to access
to a knowledge base compliant with Semantic Web standards
could potentially use Lizard’s API.
• We introduce a novel approach for automatically integrating
knowledge coming from different ontologies with a frame-driven
approach (cf. RQ4). This method aimed at finding complex
correspondences between ontology entities according the inten-
sional meaning of their models, hence abstracting from their
logical types. In this proposal, frames are considered as “unit of
meaning” [79] for ontologies and are used as a mean for repres-
enting intentional meaning of ontology entities. The frame-based
representation of entities’ meaning enables at finding complex
correspondences among entities abstracting from their logical
type thus leading a step ahead the state of the art of ontology
matching. Other potential benefits of this method related to
understanding and generating natural language will be discussed
in Section 6.3.
• This thesis introduces a component-based architecture relying
on Semantic Web standards for supporting knowledge-intensive
tasks performed by social robots, and whose design has been
guided by requirements coming from a real socially assistive
robotic application. The ultimate goal aim of the architecture
is to create a platform for easing the development of robotic
applications by providing developers with off-the-shelf software
artifacts, models and data. The strategy to pursue this goal
is to massively reuse Semantic Web technologies due to their
intrinsic availability and interoperability. Moreover, we present](https://image.slidesharecdn.com/18607690-250603002415-ccda2ab2/85/Engineering-Background-Knowledge-For-Social-Robots-Luigi-Asprino-28-320.jpg)
![Chapter 2
Background
2.1. Social Robots
Social Robots are embodied agents designed to socially interact with
people and can be categorized depending on the application domain
or depending on the type of tasks are designed to perform (for compre-
hensive overviews please refer to [29, 67, 112, 171]). Leite et al. [112]
identified four different application domains: health care, education,
work environments and public spaces, and home. Socially Assistive
Robots can be broadly classified into three categories (a review of the
assistive social robots is provided by [29, 171]): (i) Service Robots
are devices designed to support people living independently by assist-
ing them with mobility, completing household tasks, and monitoring
health and safety; (ii) Companion Robots [47] are meant to create
companionship for human beings; (iii) Coaching Robots (e.g. [60, 61])
that act as a coach to encourage human beings through a series of
therapeutic tasks for enhancing their health conditions.
2.1.1. Software Architectures for Social Robots
Despite the different application domains and the intended functions,
most of the architectures of social robots [34, 60, 61, 84, 85, 91, 95,
125, 133, 201] are constituted by the following elements:
1. A subsystem that manages the hardware devices allowing the
robot to perceive the environment (such as lasers used for nav-
igation, cameras, touch sensors, microphones etc.).
15](https://image.slidesharecdn.com/18607690-250603002415-ccda2ab2/85/Engineering-Background-Knowledge-For-Social-Robots-Luigi-Asprino-35-320.jpg)
![16 CHAPTER 2. BACKGROUND
2. A set of components dealing with the robot’s motors and actu-
ators (e.g. wheel engines, speakers).
3. A knowledge base storing information for supporting the ro-
bot’s behaviors, tracing the users’ activities or preferences, and
collecting from the operating environment (e.g. maps).
4. A multi-modal user interface that provides users with multiple
modes to interact with robots. This is typically delivered by a
voice-user interface and/or a touch-screen device.
5. A behavior controller that gathers information from perceptual
components, knowledge base and user interfaces, and decides
the next actions to perform.
6. If necessary, a supervision interface that enables to remotely
control the robot and to possibly interrupt the robot’s operation.
Using such a interface is part of one of the most common Human-
Robot Interaction experimental techniques called Wizard of
Oz [100, 168].
The architecture presented in Chapter 7 follows the structure of exist-
ing architectures and defines a subsystem that allows to dynamically
(i.e. at run-time without need of re-deploy) extend the capabilities
of the robot, thus enabling an agile and evolutionary development of
the architecture. Examples of such flexibility mechanisms can be also
found in literature. Fritsch et al. [69] proposed a flexible infrastructure
to extend the capabilities of the companion enabling the interaction
with humans. XML is used In this proposal as language for defining
the format of messages exchanged by the components and to define
the sequence of operations the robot have to perform. Similarly, the
extensible architecture introduced by Rossi et al. [173] allows to modify
and expand the multi-modal interface without impacting the rest of
the architecture. These mechanisms partially fulfill the flexibility
and extensibility requirement since does not allow to dynamically
deploy new software components. This feature is provided by our
architecture which guarantees extensibility of the robot behaviors and
capabilities. Other novel elements of our architecture concern with the
use of semantic web and linked data for managing and bootstrapping
the knowledge base.](https://image.slidesharecdn.com/18607690-250603002415-ccda2ab2/85/Engineering-Background-Knowledge-For-Social-Robots-Luigi-Asprino-36-320.jpg)
![2.2. THE SEMANTIC WEB 17
Figure 2.1: The Semantic Web stack.
2.2. The Semantic Web
The Semantic Web [18] is an extension of the Web aims at providing
a common framework that allows data to be shared and reused across
application boundaries. Standardisation for Semantic Web is under
the care of World Wide Web Consortium (W3C). The W3C standards
for the Semantic Web mainly include: XML, RDF(S), OWL and
SPARQL. Figure 2.1 shows the semantic web stack and provides an
overview of the standard technologies recommended by the W3C.
2.2.1. Extensible Markup Language (XML)
Extensible Markup Language (XML) is a markup language that defines
a set of rules for encoding documents in a both human-readable and
machine-readable format. An XML document consists of a properly
nested set of open and close tags, where each tag can have a number
of attribute-value pairs. Crucial to XML is that the vocabulary of the
tags and their allowed combinations is not fixed, but can be defined
per application of XML. In the Semantic Web context, XML is being
used as a uniform data-exchange format thus providing a common
syntax for exchange data across the web.](https://image.slidesharecdn.com/18607690-250603002415-ccda2ab2/85/Engineering-Background-Knowledge-For-Social-Robots-Luigi-Asprino-37-320.jpg)
![20 CHAPTER 2. BACKGROUND
and subdivided according to similarities and differences. While the
term ontology has been rather confined to the philosophical sphere
in the recent past, it has gained a specific role in a variety of fields
of Computer Science, such as Artificial Intelligence, Computational
Linguistics, and Database Theory and Semantic Web. In Computer
Science the term loses part of its metaphysical background and, still
keeping a general expectation that the features of the model in an
ontology should closely resemble the real world, it is referred as a
formal model consisting of a set of types, properties, and relationship
types aimed at modeling objects in a certain domain or in the world.
In early ’90s Gruber [87] gave an initial and widely accepted definition:
An ontology is a formal, explicit specification of a shared conceptual-
ization. An ontology is a description (like a formal specification of a
program) of the concepts and relationships that can formally exist for
an agent or a community of agents.
Accordingly, ontologies are used to encode a description of some world
(actual, possible, counterfactual, impossible, desired, etc.), for some
specific purpose.
In the Semantic Web, ontologies have been used as a formalism
to define the logical backbone of the Web itself. The language used
for designing ontologies in the Web of Data is the Web Ontology
Language (OWL). In the last decade there has been a lot of research for
investigating best practices for ontology design and re-use in the Web
of Data. Among the others the EU-FP7 NeOn project8
has provided
sound principles and guidelines for designing complex knowledge
networks called ontology networks. An ontology network is a set of
interconnected ontologies. According to [4], the interconnections can
be defined in a variety of ways, such as alignments, modularization
based on owl:import axioms, and versioning. Ontology networks
enable modular ontology design in which each module conceptualizes
a specific domain and can be designed by using Ontology Design
Patterns [78] and pattern-based ontology design methodologies, such
as eXtreme Desing [23].
8
http://www.neon-project.org/](https://image.slidesharecdn.com/18607690-250603002415-ccda2ab2/85/Engineering-Background-Knowledge-For-Social-Robots-Luigi-Asprino-40-320.jpg)
![2.3. ONTOLOGIES 21
2.3.1. Knowledge Management Frameworks for Social
Robots
Ontologies and Semantic Web technologies can support the develop-
ment of robotic systems and applications that deal with knowledge
representation, acquisition and reasoning. Furthermore, Semantic
Web standards enable the interlinking of local robotic knowledge with
available information and resources coming from the Web of Data.
This trend has also led to the creation of the IEEE RAS Ontologies
for Robotics and Automation Working Group (ORA WG), with the
goal of developing a core ontology and an associated methodology
for knowledge representation and reasoning in robotics and automa-
tion [161].
In this direction, different frameworks have been proposed to model,
manage and make available heterogeneous knowledge for robotic sys-
tems and applications. Focusing on service robots that operate in
indoor environments through perception, planning and action, the
ontology-based unified robot knowledge framework (OUR-K) [117]
aims at supporting robot intelligence and inference methods by in-
tegrating low-level perceptual and behavioural data with high-level
knowledge concerning objects, semantic maps, tasks, and contexts. An
ontology-based approach is also adopted in the ORO knowledge man-
agement platform [113]. The platform stores and processes knowledge
represented according to the OpenRobots Common Sense Ontology9
,
an OWL ontology based on the OpenCyc upper ontology and extended
with the definition of reference concepts for human-robot interaction.
When deployed on a robot, the knowledge base can be instantiated
with a priori common-sense knowledge and is then used as a “semantic
blackboard” where the robotic modules (such as the perception module,
the language processing module, the task planner and the execution
controller) can store the knowledge they produce and query it back.
Along the same path, research projects and initiatives, such as
KnowRob10
, RoboEarth11
and RoboBrain12
, go beyond local know-
ledge bases and, also with the emergence of cloud-based robotics, pro-
pose Web-scale approaches. KnowRob [195, 196, 197] is a knowledge
9
https://www.openrobots.org/wiki/oro-ontology
10
http://knowrob.org/
11
http://roboearth.ethz.ch/
12
http://robobrain.me/](https://image.slidesharecdn.com/18607690-250603002415-ccda2ab2/85/Engineering-Background-Knowledge-For-Social-Robots-Luigi-Asprino-41-320.jpg)
![22 CHAPTER 2. BACKGROUND
processing system and semantic framework for integrating information
from different sources, including encyclopedic knowledge, common-
sense knowledge, robot capabilities, task descriptions, environment
models, and object descriptions. Knowledge is represented and form-
ally modeled according to a reference upper ontology, defined using the
Web Ontology Language (OWL). The system supports different reas-
oning capabilities and provides interfaces for accessing and querying
the KnowRob ontology and knowledge base. Similarly, the RoboEarth
framework [198] provides a web-based knowledge base for robots to
access and share semantic representations of actions, object models
and environments, augmented with rule-based learning and reasoning
capabilities. The RoboEarth knowledge base relies on a reference
ontology, as an extension of the KnowRob ontology to (i) represent
actions and relate them in a temporal hierarchy; (ii) describe object
models to support recognition and articulation; and (iii) represent
map-based environments.
An HTTP-based API enables robots to access the knowledge
base for uploading, searching and downloading information from and
to their local knowledge bases. Along the same path, the Robo-
Brain knowledge engine [176] aims at learning and sharing knowledge
gathered from different sources and existing knowledge bases, includ-
ing linguistic resources, such as WordNet, image databases, such as
ImageNet, and Wikipedia. Although the RoboBrain knowledge base
does not explicitly adopt ontologies and Semantic Web technologies,
knowledge is represented in a graph structure and stored in a graph
database. A REST API enables robots to access RoboBrain as-a-
service, to provide and retrieve knowledge on the basis of a specific
query language.
The need to provide robots with a knowledge representation and
management framework able to handle knowledge from different
sources (including external data sources and knowledge bases) and
support multiple tasks and applications has long been considered in
robotics. However, it is only in recent years that the potential of
ontology-based knowledge representation approaches and Semantic
Web technologies has been considered to address the two aforemen-
tioned points in robotic platforms.](https://image.slidesharecdn.com/18607690-250603002415-ccda2ab2/85/Engineering-Background-Knowledge-For-Social-Robots-Luigi-Asprino-42-320.jpg)
![2.4. PATTERN-BASED ONTOLOGY DESIGN 23
2.4. Pattern-based Ontology Design
The notion of “pattern” has proved useful in design, as exemplified
in diverse areas, such as software engineering. Under the assumption
that there exist classes of problems that can be solved by applying
common solutions (as has been experienced in software engineering),
it is suggested to support reusability on the design side specifically.
To this end Ontology Design Patterns (ODPs) have been proposed
as modeling solutions to recurrent ontology design problems. ODPs
are modeling components that can be used as basic building blocks
of an ontology network. eXtreme Design (XD) is an ontology design
methodology that supports the pattern-based approach. We adopted
XD as methodology for designing the MARIO Ontology Network
presented in Chapter 3 and we extensively reused ODPs. Sections 2.4.1
and 2.4.2 briefly introduce ODPs and XD, respectively.
2.4.1. Ontology Design Patterns
Ontology Design Patterns (ODPs) [78] is an emerging technology that
favors the reuse of encoded experiences and good practices. ODPs are
modeling solutions to solve recurrent ontology design problems. They
can be of different types including: (i) logical, which typically provide
solutions for solving problems of expressivity e.g., expressing n-ary
relations in OWL; (ii) architectural, which describe the overall shape
of the ontology (either internal or external) that is convenient with
respect to a specific ontology-based task or application e.g. a certain
DL family; (iii) content, which are small ontologies that address a
specific modeling issue, and can be directly reused by importing them
in the ontology under development e.g., representing roles that people
can play during certain time periods; (iv) presentation, which provide
good practices for e.g. naming conventions.
2.4.2. eXtreme Design
eXtreme Design (XD) [163, 23, 164] is a family of methods and
associated tools, based on the application, exploitation, and definition
of ontology design patterns (ODPs) for solving ontology development
issues. XD principles are inspired by those of the agile software
methodology called eXtreme Programming (XP). The main idea of
agile software development is to be able to incorporate changes easily,](https://image.slidesharecdn.com/18607690-250603002415-ccda2ab2/85/Engineering-Background-Knowledge-For-Social-Robots-Luigi-Asprino-43-320.jpg)
![24 CHAPTER 2. BACKGROUND
Figure 2.2: An example of collected uses-story.
in any stage of the development. Instead of using a waterfall-like
method, where you first do all the analysis, then the design, the
implementation and finally the testing, the idea is to cut this process
into small pieces, each containing all those elements but only for a
very small subset of the problem. XD is test-driven, and applies the
divide-and-conquer approach as well as XP does. Also, XD adopts
pair design, as opposed to pair programming. The main principles of
the XD method can be summarised as follows:
• Customer involvement and feedback. The customer should
be involved in the ontology development and its representative
should be aware of all parts of the ontology project under devel-
opment. Interaction with the customer representative is key for
favoring the explicit expression of the domain knowledge.
• Customer stories and Competency Questions. The onto-
logy requirements and its tasks are described in terms of small
stories by the customer representative. Designers work on those
small stories and, together with the customer, transform them in
the form of Competency Questions [88] (CQs). CQs will be used
through the whole development, and their definition is a key
phase as the designers have the challenge to help the customer
in making explicit as much implicit knowledge as possible. We
asked all the partners involved in the case study to provide their
own stories. The template for providing the stories is shown](https://image.slidesharecdn.com/18607690-250603002415-ccda2ab2/85/Engineering-Background-Knowledge-For-Social-Robots-Luigi-Asprino-44-320.jpg)
![2.4. PATTERN-BASED ONTOLOGY DESIGN 25
in Figure 2.2. The fields “Partner”, “Scriber”, “e-mail” were
used for asking further clarification about the story. The Title
field helped for a better understanding the main focus of the
story. The “Priority” field was used to choose the stories to
treat first. The allowed values were High, Medium and Low.
“Depends on” allowed to specify a link between two stories. For
example, if a story was too long, it could be split into two stories
and this field allowed one to express the dependency. The last
field “Knowledge area(s)” was used for associating the story
with one or more knowledge areas which the story belonged to.
The customer stories collected together with the resulting Com-
petency Questions can be retrieved on-line13
. Other competency
questions have been extracted by analysing domain documents,
such as those used for effectuating a Comprehensive Geriatric
Assessment (CGA) of a patient.
• Content Pattern (CP) reuse and modular design. A de-
velopment project is characterised by two main sets: (i) the
problem space composed of the actual modelling issues that have
to be addressed during the project which are called “Local Use
Case” (LUC); (ii) the solution space made up of reusable model-
ling solutions, called “Global Use Case” (GUC), representing the
problem that a certain ODP provides a solution for. If there is
a CP’s GUC that matches a LUC it has to be reused, otherwise
a new module is created. An analysis of the possible strategies
for reusing CP is provided by [164].
• Collaboration and Integration. Collaboration and constant
sharing of knowledge is needed in a XD setting, in fact similar
or even the same CQs and sentences can be defined for different
stories. When this happens, it means that these stories can be
modelled by reusing a set of shared CPs.
• Task-oriented design. The focus of the design is on that part
of the domain of knowledge under investigation that is needed in
order to address the user stories, and more generally, the tasks
that the ontology is expected to address.
• Test-driven design. A new story can be treated only when
13
http://etna.istc.cnr.it/mario/D5.1/.](https://image.slidesharecdn.com/18607690-250603002415-ccda2ab2/85/Engineering-Background-Knowledge-For-Social-Robots-Luigi-Asprino-45-320.jpg)
![26 CHAPTER 2. BACKGROUND
all unit tests associated with it have been passed. An ontology
module developed for addressing a certain user story associated
to a certain competency question, is tested e.g. (i) by encoding
in the ontology a sample set of facts based on the user story, (ii)
defining one or a set of SPARQL queries that formally encode
the competency question, (iii) associating each SPARQL query
with the expected result, and (i) running the SPARQL queries
against the ontology and compare actual with expected results.
2.5. Ontology Matching
Among the various semantic technology proposed to handle heterogen-
eity Ontology Matching [181] has proved to be an effective solution
to automate integration of distributed information sources. Ontology
Matching (OM) finds correspondences between semantically related
entities of ontologies. These correspondences enable several tasks such
as ontology merging, query answering, or data translation. There
have been proposed several formalization of the matching problem, we
follow the formalization in [57] that provide a unified approach over
the previous works. The matching problem is the problem of finding
an alignment between two ontologies. An alignment is a set of 4-uple
(e1, e2, r, n) where: (i) e1 and e2 are entities defined by the first and
the second ontology, respectively; (ii) r is a relation holding between
e1 and e2, e.g., equivalence, subsumption, disjointness; (iii) n is the
confidence measuring the likelihood that the relation holds.
2.6. Linguistic Linked Open Data Resources
Many resources belonging to different domains are now being pub-
lished on-line using Linked Data principles to provide easy access to
structured data on web. This includes many linguistic resources that
are already a part of Linked Data, but made available mainly for the
purpose of being used by NLP applications.
Two of the most important linguistic linked open data resources
are WordNet [135] and FrameNet [10]. They have already been form-
alised as semantic web resources, e.g. in OntoWordNet [75], WordNet
RDF [7], FrameNet RDF [143], etc. FrameNet allows to represent
textual resources in terms of Frame Semantics. The usefulness of](https://image.slidesharecdn.com/18607690-250603002415-ccda2ab2/85/Engineering-Background-Knowledge-For-Social-Robots-Luigi-Asprino-46-320.jpg)
![2.7. COMMON SENSE KNOWLEDGE 27
FrameNet is limited by its limited coverage, and non-standard se-
mantics. An evident solution would be to establish valid links between
FrameNet and other lexical resources such as WordNet , VerbNet
and BabelNet to create wide-coverage and multi-lingual extensions of
FrameNet. By overcoming these limitations NLP-based applications
such as question answering, machine reading and understanding, etc.
would eventually be improved. Within MARIO these were important
requirements, hence we developed Framester (presented in Chapter 3):
a frame-based ontological resource acting as a hub between e.g. Fra-
meNet, WordNet, VerbNet, BabelNet, DBpedia, Yago, DOLCE-Zero,
and leveraging this wealth of links to create an interoperable predicate
space formalised according to frame semantics [65], and semiotics [70].
Data designed according to the predicates in the predicate space
created by Framester result to be more accessible and interoperable,
modulo alignments between specific entities or facts.
The closest resources to Framester are FrameBase [174] and Pre-
dicate Matrix [106]. FrameBase aimed at aligning linked data to
FrameNet frames, based on similar assumptions as Framester’s: full-
fledged formal semantics for frames, detour-based extension for frame
coverage, and rule-based lenses over linked data. However, the cov-
erage of FrameBase is limited to an automatically learnt extension
(with resulting inaccuracies) of FrameNet-WordNet mappings, and
the alignment to linked data schemas is performed manually. Anyway,
Framester could be combined with FrameBase (de)reification rules so
that the two projects can mutually benefit from their results.
Predicate Matrix is an alignment between predicates existing in
FrameNet, VerbNet, WordNet, and PropBank. It does not assume
a formal semantics, and its coverage is limited to a subset of lexical
senses from those resources. A RDF version of Predicate Matrix has
been created in order to add it to the Framester linked data cloud,
and (ongoing work) to check if those equivalences can be reused in
semantic web applications.
2.7. Common Sense Knowledge
Over the years, a number of projects aimed at generating common
sense knowledge. Regardless specific settings, the results of these
projects can be seen an ontology O =< T, A > consisting of T, a
T-box (i.e. terminology box, also called schema or vocabulary) and A](https://image.slidesharecdn.com/18607690-250603002415-ccda2ab2/85/Engineering-Background-Knowledge-For-Social-Robots-Luigi-Asprino-47-320.jpg)
![28 CHAPTER 2. BACKGROUND
is an A-box (assertion box). The outcome of these projects can be
informally classified on the basis of following criteria. (i) the process
used to generate knowledge (e.g. automatic or with humans in the
loop); (ii) the breadth of the knowledge produced (i.e. universal or
domain specific); (iii) the formalism used to represent knowledge;
(iv) the density of the knowledge (i.e. the number of assertion per
entity); (v) the richness (the variety of types and relations) and the
depth (the number of inheritance relations) of the t-box; (vi) the level
of interoperability with other datasets (i.e. the linkage of the ontology
at both intensional and extensional level); (vii) the metadata (i.e.
provenance and validity of stated facts).
Among existing projects we overview strengths, weaknesses and
results of DBpedia, ConceptNet, NELL. DBpedia14
[21] is a very
popular dataset automatically obtained from Wikipedia infoboxes.
DBpedia is the de-facto main hub of the Web of Data containing 4,58
milion entities of encyclopedic nature. However, some weaknesses of
DBpedia are the scarcity of relations among entities and the limited
depth of the vocabulary. In fact, the DBPedia ontology is induced
from Wikipedia and is only partially aligned with existing formal
theories (e.g. foundational ontologies). These drawbacks make hard
answer queries such that “what knives are used for?”, “give me all
physical entities”.
ConceptNet15
[120, 187, 186] is a large scale multilingual semantic
network that integrates knowledge from (i) Open Mind Common Sense
project [183, 184] who ran a web site that collected common sense facts
from users; (ii) existing datasets such as DBpedia, Wiktionary, Open
Multilingual WordNet, OpenCyc [115] and Umbel; (iii) Verbosity,
a game with a purpose that learns common sense knowledge from
people’s intuitive word associations. ConceptNet defines thirty lexical
and common sense relations among its entities such as: “antonym”,
“synonym”, “is used for” (that associates an object with what is used
for, e.g. “bridge” and “cross water”), “at location” (that associates
an object with its typical locations, “butter” and “refrigerator”),
“capable of” (associating an object with what it can do, e.g. “knife”
and “cut”) etc. NELL16
[35, 138] is an ongoing project aiming at
learning large semantic network (similar to ConceptNet) with a never
14
DBpedia, http://wiki.dbpedia.org/
15
ConceptNet, http://conceptnet.io/
16
NELL, http://rtw.ml.cmu.edu/rtw/](https://image.slidesharecdn.com/18607690-250603002415-ccda2ab2/85/Engineering-Background-Knowledge-For-Social-Robots-Luigi-Asprino-48-320.jpg)
![32 CHAPTER 3. AN ONTOLOGY NETWORK FOR SOCIAL ROBOTS
ledge areas that are relevant to the tasks of supporting people affected
by dementia. The knowledge areas and the ontology modules were
identified by analysing the use cases that emerged from the MARIO
project (cf. Section 1.3), These uses cases mainly describe actions
and behaviors featuring the MARIO robot. Nevertheless, they also
provide a detailed descriptions about the nature of the knowledge that
the robot has to deal with in order to behave. The MON consists of 53
modules covering 12 knowledge areas. The Ontology Network has been
developed following the eXtreme Design methodology (introduced in
Section 2.4.2) and by extensively reusing Ontology Design Patterns
(cf. Section 2.4.1). The rest of the Section is organized as follows.
Section 3.1 describes the ontology development process, Section 3.2
presents the MON’s knowledge areas and Section 3.3 outlines the most
innovative ontology modules of the Ontology Network.
3.1. Design Methodology
The MARIO Ontology Network (MON) has been designed by following
best practices and pattern-based ontology engineering methods aimed
at extensively re-using Ontology Design Patterns (ODPs) [78]. In
particular, the MON has been developed following the eXtreme Design
(XD) [23, 163] methodology (cf. Section 2.4.2). This methodology
has been extended in order to identify the knowledge areas that
are relevant to a companion robot (cf. 3.2) and to provide ontology
engineers with guidelines for re-using existing ontologies (cf. 3.1.1).
Section 3.1.2 illustrates how eXtreme Design has been configured in
for the development of MON.
3.1.1. Guidelines for Ontology Re-use
Linked Data is rapidly increasing, especially in the public sector
where opening data is becoming a consolidated institutional activity.
However, the importance of providing Linked Data with a high quality
ontology modeling is still far from being fully perceived. The result is
that Linked Data are mostly modeled by directly reusing individual
classes and properties defined in external ontologies, overlooking the
possible risks caused by such a practice. Although ontology reuse is a
recommended practice in most ontology design methodologies [182],
a standardization of ontology reuse practices is still missing. Most](https://image.slidesharecdn.com/18607690-250603002415-ccda2ab2/85/Engineering-Background-Knowledge-For-Social-Robots-Luigi-Asprino-52-320.jpg)
![3.1. DESIGN METHODOLOGY 33
Figure 3.1: The eXtreme Design workflow as extended in [164]. The
highlighted tasks involve the guidelines for ontology reuse.
literature on ontology reuse is focused on the challenging issue of
ontology selection, while our perspective is on how to implement reuse
once the selection finalized. This practice may compromise the level
of semantic interoperability that can be achieved. Therefore, the need
of clear guidelines for ontology reuse arise.
In [164] we provided a series of guidelines for ontology reuse in
the context of ontology projects that exhibit these characteristics:
(i) there is no ontology that addresses all or most of the requirements
of the local ontology project; (ii) the ontology under development
is meant to be used as a reference ontology for a certain domain;
(iii) there is the willingness to comply with existing standards. These
guidelines can be integrated into the tasks 7 and 8 of the XD workflow
(cf. Figure 3.1 and Section 2.4.2).
Ontology re-use models can be classified based on (i) the type
of reused ontology (e.g. foundational, top-level, ontology design pat-
terns, domain ontologies); (ii) the type of reused ontology fragment
(e.g. individual entities, modules, ontology design patterns, arbitrary
fragments); (iii) the amount of reused axioms (e.g. import of all
axioms, of only axioms in a given neighbourhood of an entity, of no ax-
ioms); (iv) the alignment policy (e.g. direct reuse of entities, reuse via](https://image.slidesharecdn.com/18607690-250603002415-ccda2ab2/85/Engineering-Background-Knowledge-For-Social-Robots-Luigi-Asprino-53-320.jpg)
![remembered that these families were those who have worked their
way through the first difficulties, these figures become doubly
significant.
There is, for example, a Ukrainian family from the Russian Ukraine.
It consists of the parents and four children between the ages of
three and fifteen. Ever since the family came to the United States
they have had one or more lodgers to help them pay the rent. At
present they have three men paying four dollars a month each; and
as the father, who had been working in the stockyards for nineteen
dollars a week, was discharged two months ago, the wife has been
working in a spring factory to support the family.
Then there is a Polish family, composed of the parents and four
children under fourteen, two of them children of the man by a
former wife. The father has been in this country since 1894, but his
wife has been here only since 1910. For two years after their
marriage the wife worked at night, scrubbing from 6.30 to 9.30 p.m.,
and received twenty-four dollars a month. Then there was an
interval while her children were babies, during which she did not
work, but the family lived on the earnings of the father. For the last
two years, however, his work has been slack, first because of a
strike, and later due to an industrial depression in his trade, and the
mother is again at work, this time in a tailor shop, earning ten
dollars a week.
The effect of the mother's work in decreasing the child's chances for
life has been made clear by the studies of the Children's Bureau in
Johnstown,[10] Montclair,[11] and Manchester,[12] in all of which a
higher rate of infant mortality was shown for children of mothers
gainfully employed.
The effect of the mother's work on the family relationship and the
home life of the family group is, of course, not measurable in
absolute terms. The leaders of the various national groups, however,
have repeatedly emphasized the fact that the absence of the women
from the home has created entirely new problems in the family life.](https://image.slidesharecdn.com/18607690-250603002415-ccda2ab2/85/Engineering-Background-Knowledge-For-Social-Robots-Luigi-Asprino-56-320.jpg)
![the open air. When an attempt is made to do this in almost any large
city the tomatoes get not only the sunshine, but the soot and dirt of
the city. The more particular Italians here will not make tomato
paste outdoors, but large numbers of Italian families continue to
make it, as can be seen by a walk through any Italian district in late
August or early September.
In general, in the groups in which cooking was highly developed, a
great deal of time was devoted to the preparation of food. If the
housewife wishes to reduce her work in this country, she finds that
some of the ingredients which make our cooking simpler are
unknown to her. The Bohemians, for example, do not know how to
use baking powder, and the same is true of the women in
Lithuanian, Polish, and Russian groups, where the art of cooking is
less developed.
With this lack of experience in housekeeping under comparable
conditions, the foreign-born housewife finds the transition to
housekeeping in this country difficult at best. As a matter of fact,
however, the circumstances under which she must make the change
are often of the worst. She is expected to maintain standards of
cleanliness and sanitary housekeeping that have developed with
modern systems of plumbing and facilities for the disposal of waste
that are not always to be found in the districts in which she lives.
Even a skillful housewife finds housekeeping difficult in such houses
as are usually occupied by recently arrived immigrants.
WATER SUPPLY ESSENTIAL
In the first place, there is the question of water supply. Cleanliness
of house, clothing, and even of person is extremely difficult in a
modern industrial community, without an adequate supply of hot and
cold water within the dwelling. We are, however, very far from
realizing this condition. In some cities[13] the law requires that there
shall be a sink with running water in every dwelling, but in other
cities even this minimum is not required. The United States
Immigration Commission, for example, found that 1,413 households](https://image.slidesharecdn.com/18607690-250603002415-ccda2ab2/85/Engineering-Background-Knowledge-For-Social-Robots-Luigi-Asprino-69-320.jpg)
![out of 8,651 foreign-born households studied in seven large cities,
shared their water supply with other families. Conditions have
improved in this respect during the last decade, but it is a great
handicap to efficient housekeeping if water has to be carried any
distance. Further inconvenience results if running hot water is not
available, which is too often the case in the homes of the foreign
born.
THIS PUMP SUPPLIES WATER TO FOUR FAMILIES
Cleanliness is also dependent, in part, upon the facilities for the
disposition of human waste, the convenient and accessible toilet
connected with a sewer system. These facilities are lacking in many
immigrant neighborhoods, as has been repeatedly shown in various
housing investigations. For example, in a Slovak district in the
Twentieth Ward, Chicago, 80 per cent of the families were using
toilets located in the cellar, yard, or under the sidewalk, and in many
cases sharing such toilets with other families. One yard toilet was
used by five families, consisting of twenty-eight persons.[14] The
danger to health, and the lack of privacy, that such toilet
accommodations mean have been often emphasized. In addition, it](https://image.slidesharecdn.com/18607690-250603002415-ccda2ab2/85/Engineering-Background-Knowledge-For-Social-Robots-Luigi-Asprino-70-320.jpg)
![Yet there is abundant evidence that many families have had to live in
just such conditions.
The housework for the foreign-born housewife is often complicated
by other factors. One is the practice to which reference has been
already made of taking lodgers to supplement the father's wages. In
discussing this subject from the point of view of the lodger, it has
been pointed out that the practice with reference to the taking of
boarders and lodgers varies in different places and among different
groups. The amounts paid were not noted there, but they become
important when considered together with the service asked of the
housewife. Usually the boarder or lodger pays a fixed monthly sum—
from $2 to $3.50, or, more rarely, $4 a month—for lodging, cleaning,
washing, and cooking; his food is secured separately, the account
being entered in a grocery book and settled at regular intervals.
Sometimes the lodger does his own buying, but the more common
custom is to have the housewife do it. Occasionally he does his own
cooking, in which case payment for lodging secures him the right to
use the stove. More rarely, as in some of the Mexican families visited
in Chicago in 1919, he is a regular boarder, paying a weekly sum for
room and board.
Just what keeping lodgers means in adding to the duties of the
housewife can be seen from the following description of the work of
the Serbo-Croatian women in Johnstown, Pennsylvania:[15]
The wife, without extra charge, makes up the beds, does the
washing and ironing, and buys and prepares the food for all the
lodgers. Usually she gets everything on credit, and the lodgers
pay their respective shares biweekly. These conditions exist to
some extent among other foreigners, but are not so prevalent
among other nationalities in Johnstown as among the Serbo-
Croatians.
In a workingman's family, it is sometimes said, the woman's
working day is two hours longer than the man's. But if this](https://image.slidesharecdn.com/18607690-250603002415-ccda2ab2/85/Engineering-Background-Knowledge-For-Social-Robots-Luigi-Asprino-72-320.jpg)
![late at night at laundry or cleaning; 49 women, for example, washed
in the evening; 25 washed either Saturday, Sunday, or evenings.
HOUSING IMPROVEMENT
Enough has probably been said to show that the work of caring for
the house under the conditions existing in most immigrant
neighborhoods, is unnecessarily difficult for the foreign-born
housewife. The most obvious point at which these burdens might be
lightened so that the housewife could have time for other duties,
would appear to be through improvement of housing. With an
awakened realization of this fact, both on the part of the foreign-
born woman herself and the community of which she is an inevitable
part, will come the solution of these difficulties. A protest, however
inarticulate or indirectly expressed by her, will find its response in a
growing realization that plans for improvement must be developed.
The several housing projects that have already been offered are
suggestive of the problems and possibilities along this line rather
than useful as hard-and-fast solutions. They not only meet the needs
of the more inadequate immigrant housing conditions, but provide
improvement upon most native-born conditions. In this connection
interest naturally centers on the war-time housing projects of the
United States government, on the experiment of the Massachusetts
Homestead Commission at Lowell, and on certain enterprises carried
out by so-called limited dividend companies. The first two are
especially interesting, in that they recognize that supplying houses to
the workers is not a function that can be wholly left to private
initiative.
It is not possible to discuss these projects in detail, nor is it
necessary.[16] It is sufficient to consider them here with reference to
the contributions they might make in helping the immigrant
housewife. In the first place, they provide for a toilet and a bath in
every house, and a supply of running water that is both adequate
and convenient. In the matter of kitchen equipment there is an
attempt to provide some of the conveniences. Both provide a sink](https://image.slidesharecdn.com/18607690-250603002415-ccda2ab2/85/Engineering-Background-Knowledge-For-Social-Robots-Luigi-Asprino-74-320.jpg)
Engineering Background Knowledge For Social Robots Luigi Asprino
- 1. Engineering Background Knowledge For Social Robots Luigi Asprino download https://ebookbell.com/product/engineering-background-knowledge- for-social-robots-luigi-asprino-37215380 Explore and download more ebooks at ebookbell.com
- 2. Here are some recommended products that we believe you will be interested in. You can click the link to download. Bounding Uncertainty In Civil Engineering Theoretical Background 1st Edition Alberto Bernardini https://ebookbell.com/product/bounding-uncertainty-in-civil- engineering-theoretical-background-1st-edition-alberto- bernardini-2030302 Creative Engineering Design Assessment Background Directions Manual Scoring Guide And Uses 1st Edition Christine Charyton Auth https://ebookbell.com/product/creative-engineering-design-assessment- background-directions-manual-scoring-guide-and-uses-1st-edition- christine-charyton-auth-4342540 Process Engineering Renewal 1 Background And Training Jeanclaude Andr https://ebookbell.com/product/process-engineering- renewal-1-background-and-training-jeanclaude-andr-10911678 Process Engineering Renewal 1 Background And Training Andr https://ebookbell.com/product/process-engineering- renewal-1-background-and-training-andr-12075140
- 3. Engineering Biostatistics An Introduction Using Matlab And Winbugs 1st Edition Brani Vidakovic https://ebookbell.com/product/engineering-biostatistics-an- introduction-using-matlab-and-winbugs-1st-edition-brani- vidakovic-44977548 Engineering For Sustainable Future Selected Papers Of The 18th International Conference On Global Research And Education Interacademia 2019 1st Edition Annamria R Vrkonyikczy https://ebookbell.com/product/engineering-for-sustainable-future- selected-papers-of-the-18th-international-conference-on-global- research-and-education-interacademia-2019-1st-edition-annamria-r- vrkonyikczy-45687454 Engineering Mechanics 15th Edition Russell C Hibbeler https://ebookbell.com/product/engineering-mechanics-15th-edition- russell-c-hibbeler-46084234 Engineering A Compiler 3rd Edition Keith D Cooper Linda Torczon https://ebookbell.com/product/engineering-a-compiler-3rd-edition- keith-d-cooper-linda-torczon-46123798 Engineering Electromagnetics A Collection Of Theorems Principles And Field Representations Ramakrishna Janaswamy https://ebookbell.com/product/engineering-electromagnetics-a- collection-of-theorems-principles-and-field-representations- ramakrishna-janaswamy-46124746
- 7. ENGINEERING BACKGROUND KNOWLEDGE FOR SOCIAL ROBOTS
- 8. Studies on the Semantic Web Semantic Web has grown into a mature field of research. Its methods find innovative applica- tions on and off the World Wide Web. Its underlying technologies have significant impact on adjacent fields of research and on industrial applications. This book series reports on the state of the art in foundations, methods, and applications of Semantic Web and its underlying tech- nologies. It is a central forum for the communication of recent developments and comprises research monographs, textbooks and edited volumes on all topics related to the Semantic Web. Editor-in-Chief: Prof. Dr. Pascal Hitzler Department of Computer Science, Kansas State University, Manhattan, KS 66502, USA Email: hitzler@k-state.edu Editorial Board: Diego Calvanese, Vinary Chaudhri, Fabio Ciravegna, Michel Dumontier, Dieter Fensel, Fausto Giunchiglia, Carole Goble, Asunción Gómez Pérez, Frank van Harmelen, Manfred Hauswirth, Ian Horrocks, Krzysztof Janowicz, Michael Kifer, Riichiro Mizoguchi, Mark Musen, Daniel Schwabe, Barry Smith, Steffen Staab, Rudi Studer and Elena Simperl Volume 048 Previously published in this series: Vol. 047 Ilaria Tiddi, Freddy Lécué, Pascal Hitzler (Eds.), Knowledge Graphs for Explainable Artificial Intelligence: Foundations, Applications and Challenges Vol. 046 Daniel Dominik Janke, Study on Data Placement Strategies in Distributed RDF Stores Vol. 045 Pavlos Vougiouklis, Neural Generation of Textual Summaries from Knowledge Base Triples Vol. 044 Diego Collarana, Strategies and Techniques for Federated Semantic Knowledge Integration and Retrieval Vol. 043 Filip Ilievski, Identity of Long-Tail Entities in Text Vol. 042 Fariz Darari, Managing and Consuming Completeness Information for RDF Data Sources Vol. 041 Steffen Thoma, Multi-Modal Data Fusion Based on Embeddings Vol. 040 Marilena Daquino, Mining Authoritativeness in Art Historical Photo Archives. Semantic Web Applications for Connoisseurship Vol. 039 Bo Yan, Geographic Knowledge Graph Summarization Vol. 038 Petar Ristoski, Exploiting Semantic Web Knowledge Graphs in Data Mining Vol. 037 Maribel Acosta Deibe, Query Processing over Graph-structured Data on the Web Vol. 036 E. Demidova, A.J. Zaveri, E. Simperl (Eds.), Emerging Topics in Semantic Technologies Vol. 035 Giuseppe Cota, Inference and Learning Systems for Uncertain Relational Data Vol. 034 Ilaria Tiddi, Explaining Data Patterns using Knowledge from the Web of Data Vol. 033 Anne E. Thessen, Application of Semantic Technology in Biodiversity Science Vol. 032 Pascal Hitzler et al. (Eds.), Advances in Ontology Design and Patterns Vol. 031 Michael Färber, Semantic Search for Novel Information Vol. 030 Hassan Saif, Semantic Sentiment Analysis in Social Streams Vol. 029 A. Ławrynowicz, Semantic Data Mining: An Ontology-Based Approach Vol. 028 R. Zese, Probabilistic Semantic Web: Reasoning and Learning ISSN 1868-1158 (print) ISSN 2215-0870 (online)
- 9. ENGINEERING BACKGROUND KNOWLEDGE FOR SOCIAL ROBOTS Luigi Asprino University of Bologna, Bologna, Italy
- 10. © 2020 Akademische Verlagsgesellschaft AKA GmbH, Berlin All rights reserved. No part of this book may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, without prior written permission from the publisher. ISBN 978-3-89838-757-6 (AKA, print) ISBN 978-1-64368-108-5 (IOS Press, print) ISBN 978-1-64368-109-2 (IOS Press, online) doi: 10.3233/SSW48 Bibliographic information available from the Katalog der Deutschen Nationalbibliothek (German National Library Catalogue) at https://www.dnb.de Dissertation, approved by the University of Bologna Date of the defense: 3 April 2019 Supervisors: Paolo Ciancarini and Valentina Presutti Publisher Akademische Verlagsgesellschaft AKA GmbH, Berlin Represented by Co-Publisher IOS Press IOS Press BV Nieuwe Hemweg 6B 1013 BG Amsterdam The Netherlands Tel: +31 20 688 3355 Fax: +31 20 687 0019 email: order@iospress.nl LEGAL NOTICE The publisher is not responsible for the use which might be made of the following information. PRINTED IN THE NETHERLANDS
- 11. To Serena
- 13. Abstract Social robots are embodied agents that continuously perform knowl- edge-intensive tasks involving several kinds of information coming from different heterogeneous sources. Providing a framework for engineer- ing robots’ knowledge raises several problems like identifying sources of information and modeling solutions suitable for robots’ activities, integrating knowledge coming from different sources, evolving this knowledge with information learned during robots’ activities, ground- ing perceptions on robots’ knowledge, assessing robots’ knowledge with respect humans’ one and so on. Semantic Web research has faced with most of these issues and can provide robots with the means for creating, organizing, querying and reasoning over knowledge. In fact, Semantic Web standards allow to easily integrate data generated by a variety of components, thus enabling robots to make decisions by taking into account knowledge about physical world, data coming from their operating environment, information about social norms, users’ preferences and so on. Semantic Web technologies provide flexible solutions that allow to extend and evolve robots’ knowledge over time. Linked (Open) Data paradigm (a result of research in the Semantic Web field) lets to easily reuse (i.e. integrate with robots’ knowledge) existing external datasets so to bootstrap a robot’s knowledge base with relevant information for its activities. Linked Data also provides a mechanism that allows robots to mutually share knowledge. Existing solutions for managing robots’ knowledge only partially exploit the potential of Semantic Web technologies and Linked Data. This thesis introduces a component-based architecture relying on Semantic Web standards for supporting knowledge-intensive tasks performed by so- cial robots, and whose design has been guided by requirements coming from a real socially assistive robotic application. All the components contribute to and benefit from the knowledge base which is the corner- vii
- 14. viii ABSTRACT stone of the architecture. The knowledge base is structured by a set of interconnected and modularized ontologies which are meant to model information relevant for supporting robots in their daily activities. The knowledge base is originally populated with linguistic, ontological and factual knowledge retrieved from the Linked Open Data. The access to the knowledge base is guaranteed by Lizard, a tool that provides software components with an API for accessing facts stored in the knowledge base in a programmatic and object-oriented way. This thesis also introduces two methods for engineering knowledge needed by robots: (i) A novel method for automatically integrating knowledge coming from heterogeneous sources with a frame-driven approach. (ii) A novel empirical method for assessing foundational distinctions over Linked Open Data entities from a common sense perspective (e.g. deciding if an entity inherently represents a class or an instance from a common sense perspective). These methods realize two tasks of a more general procedure meant to automatically evolve robots’ knowledge by automatically integrating information coming from heterogeneous sources, and to generate common sense knowledge by using Linked Open Data as empirical basis. Feasib- ility and benefits of this architecture have been assessed through a prototype deployed in a real socially assistive scenario, whose this thesis presents two applications and the results of a qualitative and quantitative evaluation.
- 15. Contents Chapter 1. Introduction 1 Chapter 2. Background 15 Chapter 3. An Ontology Network for Social Robots in Assistive Context 31 Chapter 4. Providing Linked Open Data as Background Knowledge for Social Robots 77 Chapter 5. Accessing Background Knowledge using Liz- ard 107 Chapter 6. A Frame-based Approach for Integrating Ontologies 137 Chapter 7. A Knowledge Base Centered Software Ar- chitecture for Social Robots 149 Chapter 8. Conclusion and Future Work 177 Appendix A. Code Generated by Lizard 185 ix
- 17. List of Figures 1.1 The Kompaı̈-2 robot and its user interface. 10 2.1 The Semantic Web stack. 17 2.2 An example of collected uses-story. 24 3.1 The eXtreme Design workflow as extended in [164]. The highlighted tasks involve the guidelines for ontology reuse. 33 3.2 The template of the user stories provided by the cus- tomer representatives. 38 3.3 The network of ontologies constituting the MARIO Ontology Network. 39 3.4 The diagram of the Affordance ontology. 46 3.5 Two equivalent action-selection schemes. 49 3.6 The UML class diagram of CGA ontology. 53 3.7 The UML class diagram of the Co-Habitation status ontology. 55 3.8 The UML class diagram of the Medication Use ontology. 56 3.9 The UML class diagram of the Capability Assessment ontology. 57 3.10 The UML class diagram of the SPMSQ ontology. 59 3.11 The UML class diagram of the ESS ontology. 60 3.12 The UML class diagram of the CIRS ontology. 62 3.13 The UML class diagram of the MNA ontology. 62 3.14 The UML class diagram of tagging ontology. 64 3.15 The UML class diagram of Music ontology. 70 4.1 An example of Framester representation for the concept G suit. 80 xi
- 18. xii LIST OF FIGURES 4.2 SENECA approach for assessing whether a DBpedia entity is a class or an instance (Figure 4.2a) and whether it is a physical object or not (Figure 4.2b). 92 5.1 A diagram that shows the intuition behind Lizard and its operating scenario. 111 5.2 The solution stack provided by Lizard that allow ap- plications to interact with a Knowledge Base. 112 5.3 An example showing how the hierarchy of classes defined in the input ontology (Figure 5.3a) is reflected in the Java classes generated by Lizard (Figure 5.3b). 119 5.4 The Action ontology module of the Mario Ontology Network. 120 5.5 A simple ontology arising a name clash in the method signatures. 124 6.1 The UML class diagram of the Ontology Design Pattern Participation. 141 6.2 An example of alignment between the object property isParticipantIn and the frame Competition. A dashed line represents a possible correspondence between ele- ments of the two models. These edges are labeled with a confidence measure based on the semantic text similarity of the two elements. 144 6.3 The workflow summarizing the macro steps of the pro- posed approach for matching two ontologies. 146 7.1 The software architecture of the social robot. 154 7.2 Architectural model of the CGA and Reminiscence applications 162 7.3 Example of prompting questions formulation from user- specific knowledge graph 169
- 19. List of Tables 3.1 Advantages and Disadvantages of different approaches to ontology reuse. 35 3.1 Advantages and Disadvantages of different approaches to ontology reuse. 36 3.2 Competency questions answered by the Affordance ODP. 46 3.3 Ontology modules imported/reused by the CGA onto- logy. 53 3.4 Competency questions answered through the Co-Habi- tation status ontology. 56 3.5 Competency questions answered through the Medica- tion Use ontology. 57 3.6 Competency questions answered through the Capability Assessment ontology. 58 3.7 Competency questions answered through the SPMSQ ontology. 59 3.8 Competency questions answered through the ESS onto- logy. 61 3.9 Competency questions answered through the CIRS on- tology. 61 3.10 Competency questions answered through the Tagging ontology. 64 3.11 Ontology modules imported/reused by the Tagging ontology. 65 4.1 CIC dataset crowd-based annotated dataset of classes and instances. The table provides an insight of the data- set per level of agreement. Agreement values computed according to Formula 4.1. 97 xiii
- 20. xiv LIST OF TABLES 4.2 POC dataset: crowd-based annotated dataset of phys- ical objects. The table provides an insight of the dataset per level of agreement. Agreement values computed according to Formula 4.1. 98 4.3 Results of SENECA on the Class vs. Instance and Physical Object classifications compared against the reference datasets described in Section 4.2.5. P* , R* and F* 1 indicate precision, recall and F1 measure on Class (C), Instance (I), Physical Object (PO) and complement of Physical Object (NPO). F1 is the average of the F1 measures. 99 4.4 Results of the Support Vector Machine classifier on Class vs. Instance task against the reference datasets described in Section 4.2.5. The first four columns indic- ate the features used by the classifier: A is the abstract, U is the URI, E are incoming and outgoing properties, D are the results of the alignment-based methods. P* , R* , F* 1 indicate precision, recall and F1 measure on Class (C) and Instance (I). F1 is the average of the F1 measures. 101 4.5 Results of the Support Vector Machine classifier on Physical Object classification task against the refer- ence datasets described in Section 4.2.5. The first four columns indicate the features used by the classifier: A is the abstract, U is the URI, E are incoming and outgo- ing properties, D are the results of the alignment-based methods. P* , R* , F* 1 indicate precision, recall and F1 measure on Physical Object (PO) and the complement of Physical Object (NPO). F1 is the average of the F1 measures. 102 6.1 An example of association ontology entity-frames. 142
- 21. Chapter 1 Introduction Social robots [12, 27, 46, 54, 51, 67] are autonomous embodied agents that interact, collaborate, communicate with humans, by following the behavioral norms expected by people with whom robots are in- tended to interact. Several definitions have been proposed for the term “social robot”, but all of them broadly agree that a social ro- bot has the following characteristics: (i) Physical embodiment, i.e. a social robot has a physical body; (ii) Sociality, i.e. a social robot is able to interact with people by showing human-like features while following the social rules (defined through society) attached to its role; (iii) Autonomy, i.e. a social robot makes decisions by itself (the autonomy is sometimes limited in testing phase, like in the Wizard of Oz experimental setting [100, 168]). In recent years, the field of socially assistive robotics [62] has emerged given the great potential of social robots in supporting people with cognitive impairment or physical disability [17, 86, 94, 139]. As overviewed in [127], existing robotic technologies for care range from pet-like devices to advanced anthropomorphic mobile robotic assistants. While service robots often focus on providing physical support, a socially assistive robot aims to provide cognitive support through social interaction. In order to effectively interact, communicate, collaborate with humans, robots should demonstrate some intelligence. Generally, the approaches for building such an intelligence can be classified into two categories, namely, symbolic and subsymbolic approaches [159, 185]. In symbolic approaches real-world entities, facts, rules and concepts are formalized by means of symbols. Symbols are stored in a knowledge base and are manipulated to make conclusions and take decisions. 1
- 22. 2 CHAPTER 1. INTRODUCTION Conversely, subsymbolic approaches attempt to address problems without building an explicit representation of concepts and entities. An example of subsymbolic approaches are artificial neural networks, i.e. large networks of extremely simple numerical processors (i.e. neurons), massively interconnected and running in parallel. These networks consume and produce numerical vectors. The connections between neurons determine how input vectors is transformed to an output vector. In these systems, knowledge is not encoded with symbols but rather in the pattern of numerical strengths of the connections between neurons. These networks can “learn” to perform a given task by seeing a set of examples. An example is a pair of a possible input for the task at hand with (possibly) the corresponding desirable output. From these examples a neural network can learn the optimal weights to predict the desirable output from a given input. There are benefits and drawbacks of both kinds of approaches. While subsymbolic approaches achieve better performance in specific tasks, they need a large quantity of examples to learn optimal strengths. These examples could be hardly available for certain tasks. The other main issue with subsymbolic systems is the inability of explaining why a certain output is provided for a given input. This could be a non-negligible problem for many domain, such as medicine. In contrast, symbolic systems are able to explain their decisions and do not require examples. However, these systems need considerable effort for defining symbols and designing rules and methods to manipulate symbols for solving problems. Integrating the two approaches have also become more common to benefit of both strategies (e.g. [82, 89, 134]). The framework proposed in this thesis benefits of both symbolic and subsymbolic techniques. Subsymbolic techniques are used in perceptual tasks (such as translation of spoken language into text), whereas symbolic techniques are used for controlling the robot at an higher level [83]. In particular, subsymbolic subsystems of the robot transform low-level perception in symbols so to enable the symbolic processing of the control system. In this way the framework benefits of the state-of-the- art performance on perceptual tasks of subsymbolic techniques without compromising the possibility of having a system that is deterministic and able to explain its behavior and decisions (important requirements for the case study of this thesis, cf. Section 1.3). In order to show human-like feature the robot should be able to
- 23. 3 manipulate a human-like set of symbols, called background knowledge. This could be informally defined as the knowledge that a robot need in order to operate. This background knowledge includes (but it is not limited to): linguistic, encyclopedic and procedural knowledge as well as knowledge concerning the physical world and social norms. All of these kind of knowledge are involved in a potential human-robot interaction. For example, suppose that a person asks a robot to “cut a slice of bread”. In order to fulfill this request, the robot should resort to: (i) linguistic knowledge in order to understand what the person says (e.g. to associate the word “cut” to the meaning “detaching with a sharp-edged instrument”); (ii) encyclopedic knowledge in order to figure out what are the entities involved in the request (e.g. what is a a slice of bread); (iii) procedural knowledge in order to realize the steps to undertake (e.g. take the bread knife, put the bread on a cutting board etc.); (iv) physical world knowledge in order to figure out where the entities involved in the request are usually located, and if, what the person asks, is feasible and is something that the robot can do (e.g. knives and bread can be usually found in a kitchen); (v) social norms knowledge in order to check if, what the person asks, is something socially acceptable (i.e. cutting a slice of bread is acceptable but stabbing someone is generally immoral). This simple example shows the complexity and the need of these types of knowledge. Once the robot receives a user request, it is expected to interpret it, namely, it needs to associate the user request with an internal representation of its meaning. This representation should use referents with a formal semantics for the robot. From these referents the robot can access the other kinds of knowledge. For example from a referent representing the meaning of the word “cut”, i.e. “detaching with a sharp-edged instrument”, the robot is be able to access to procedural knowledge to figure out how to cut something. This is possible if all these types of knowledge are connected, well-organized and available to the robot. Providing such a framework for engineering robots’ knowledge raises several problems like identifying sources and modeling informa- tion relevant for robots’ activities, integrating knowledge coming from different sources, evolving this knowledge with information learned during robots’ activities, grounding perceptions on robots’ knowledge, assessing robots’ knowledge with respect humans’ one and so on. Sev- eral knowledge representation approaches for robots can be found in literature, we present the most relevant that are employed in robotic
- 24. 4 CHAPTER 1. INTRODUCTION architectures. The choice of a knowledge representation formalism has an impact on the expressiveness (the breadth of concepts that can be represented), on the ability infer logical consequences form asserted facts in a tractable manner (i.e. tractability) and on the amount of tools and knowledge already available that can be provided to robots for supporting their tasks. Predicate logic is a collection of formal systems that includes: pro- positional logic (dealing with zero-arity predicates, called propositions), first-order logic (FOL) (dealing with terms of higher arity) and higher order logic (in which predicates can be arguments of other predicates). Unlike propositional logic, first-order logic and higher order logic are undecidable (i.e. it does not exist a method for validating all the formulas). To overcome the undecidability of FOL, some fragments of the logic have been proposed (e.g. horn clause and description logic). Propositional logic, horn clause (which constitutes the foundation of logic programming languages), description logic [8] (at the basis of the W3C’s OWL-DL standard) are used in several robotic frame- works (e.g. [131, 142, 176, 196]). Modal logic extends propositional and predicate logic to include the modality operator. A modal is an operator that qualifies statements (e.g. usually, possibly). An example of robotic framework based on modal logic is Tino [50]. Temporal logic allows to qualify proposition in terms of time (e.g. “I am always hungry”). It is used in some robotic frameworks such as [59, 104]. Another logics particularly relevant for robotic frameworks (e.g. [111]) is the probabilistic logic which provides a formalism able to handle uncertainty. All of these logic frameworks allow to potentially encode robot’s knowledge and are supported by several off-the-shelf reasoning tools. Although some of them show high expressive power and attractive computational features, they almost lack of existing resources that could provide robots with a comprehensive background knowledge for their tasks. The choice of a framework impacts on the effort required to provide a robot with a considerable amount of knowledge for its tasks. This problem is even more significant for social robots which need of a sizable and heterogeneous knowledge bases to operate. The Semantic Web [18] standards give a good trade-off among the expressiveness, tractability and availability of resources and tools for manipulating such knowledge. The Semantic Web (cf. Section 2.2) is an extension of the World Wide Web aimed at providing a framework
- 25. 1.1. GOALS OF THE THESIS 5 that allows data to be shared with a common syntax and semantics. It based on a language defining the syntax of data to be shared (i.e. XML), a model defining the format of data (i.e. RDF) and a language to formally specify the semantics of data (i.e. OWL). OWL is derived from description logics and comes in three forms: (i) OWL Full for maximum expressiveness, but undecidable; (ii) OWL DL designed to provide the maximum expressiveness while retaining decidability; (iii) OWL Lite that supports taxonomies and simpler constraints with a limited expressiveness but attractive computational features. Se- mantic Web standards provide a good expressive power (equivalent to Description Logics), without compromising decidability and tractabil- ity. It is supported by a pletora of off-the-shelf reasoners, knowledge management systems, as well as tools for creating, organizing and integrating knowledge. Another benefit of implementing a framework that relies on Semantic Web technologies is the opportunity of ex- ploiting knowledge available as Linked Open Data (LOD). LOD is a huge web of data (∼200 billion linked facts1 ) formally (and uniformly) represented in RDF and OWL, and openly available on the Web. 1.1. Goals of the Thesis The aim of this thesis can be summarized in the following question. RQ0: To what extent Semantic Web technologies and Linked Data can be used to create, organize, access to, and evolve robot’s background knowledge? Existing solutions for managing robots’ knowledge (such as RoboE- arth [198], RoboBrain [176], ORO [113], ORA [161], RACE [172], and OUR-K framework [117]) only partially exploit the potential of Semantic Web technologies and Linked Data. In these projects, Semantic Web technologies are mostly employed to address syntactic heterogeneity of data, to define conceptualizations of the robots’ knowledge, and, more rarely, to include external datasets within the robots’ knowledge base. Most of these frameworks focused on a single dimension of knowledge while disregarding more social aspects. In fact knowledge manipulated by these frameworks are often related to the interaction of the robot with the physical environment for supporting navigation [142], inter-robot communication [198], manipulation of 1 LODCloud, https://lod-cloud.net/
- 26. 6 CHAPTER 1. INTRODUCTION the objects [196], representation of robot’s actions [199]. We claim that robots’ architectures can profoundly benefit of Se- mantic Web technologies and Linked Data paradigm. (i) Semantic Web technologies could enable an incremental and iterative develop- ment of the architecture. (ii) Semantic Web standards allow to easily integrate data generated by a variety of components, thus enabling robots to make decisions by taking into account knowledge about the physical world, data coming from their operating environment, inform- ation about social norms, users’ preferences and so on. (iii) Semantic Web technologies provide flexible solutions to extend and evolve robots’ knowledge over time. (iv) Linked (Open) Data paradigm lets to easily reuse (i.e. integrate with robots’ knowledge) existing external datasets so to bootstrap knowledge base with relevant information for robots’ activities. (v) Linked Data also provides a mechanism that allows robots to mutually share knowledge. As anticipated with the main research question RQ0, the goal of this thesis is to investigate feasibility and benefits of engineering background knowledge of social robots with a framework based on Semantic Web technologies and Linked Data. This investigation has been carried out in a socially assistive context (presented in Section 1.3) by following the methodology described in Section 1.4. The research question RQ0 can be decomposed into the following sub-questions which will be investigated and discussed in the following chapters (one question for each Chapter, from Chapter 3 to Chapter 7). RQ1: What kind of knowledge a robot needs to operate in socially assistive context? What exiting ontologies can be used to or- ganize the robot’s knowledge? What ontologies need to be advanced? What domains of interest in this context miss of a conceptualization? RQ2: What Linked Data can provide background knowledge for social robots tasks? RQ3: How to provide robots with access to knowledge? RQ4: How to integrate robot’s knowledge with data coming from robot’s experience? RQ5: How Semantic Web technologies can be orchestrated to support robot tasks?
- 27. 1.2. CONTRIBUTIONS OF THE THESIS 7 1.2. Contributions of the Thesis This thesis contributes to goals presented in Section 1.1 as follows. • We have devised a set of interconnected and modularised onto- logies, i.e. the MARIO Ontology Network (MON), which are meant to model all knowledge areas that are relevant for robots’ activities in socially assistive contexts (cf. RQ1). This ontology network defines reference models for representing and structur- ing the knowledge processed by the robot. MON provides a robot with the means for creating, organizing, querying and reasoning over a background knowledge. MON reuses and integ- rates state-of-the-art ontologies in various domains (such those related to personal information, social and multimedia contents), and, proposes novel solutions in medical domain (e.g. CGA Ontology, cf. Section 3.3.2) and in robotic domain (i.e. Afford- ance Ontology, cf. Section 3.3.1). These modules enables robots to assess the medical, psycho-social and functional status of a person and to decide the most appropriate action to perform in a given situation. • The knowledge base is originally populated with lexical, linguistic and factual knowledge retrieved from Linked Open Data. The thesis presents a novel process for generating, integrating and assessing this knowledge (cf. RQ2). Moreover, we propose a novel empirical method for assessing foundational distinctions over Linked Open Data entities from a common sense perspective (e.g. deciding if an entity inherently represents a class or an instance from a common sense perspective). This method realizes the first step of a more general procedure meant to automatically generate common sense knowledge from Linked Open Data (cf. RQ2). These methods advance state-of-the-art in Semantic Web by proposing standardized techniques for creating an integrated repository of linguistic, factual, encyclopedic, ontological and common sense knowledge. The benefits of these techniques as well as the resulting datasets are not limited to robotic domain, but also extend to every application domain that requires a rich knowledge base to operate [49]. • We have developed an object-RDF mapper (called Lizard) that facilitates software components to interact with an RDF know-
- 28. 8 CHAPTER 1. INTRODUCTION ledge base (cf. RQ3). In particular, given an RDF knowledge base and an OWL ontology describing its structure, Lizard provides applications with an API for accessing RDF facts stored in a knowledge base following the object-oriented paradigm. The API reflects the semantics of the input ontology and al- lows transparent access to the knowledge base, Differently from existing systems, Lizard exposes the API following the REST architectural style over HTTP. This tool is aimed at easing the software development of knowledge-aware systems by filling the gap between Semantic Web technologies and Object-Oriented applications. The benefits of Lizard are not to be intended only for robotic domain, but every application that need to access to a knowledge base compliant with Semantic Web standards could potentially use Lizard’s API. • We introduce a novel approach for automatically integrating knowledge coming from different ontologies with a frame-driven approach (cf. RQ4). This method aimed at finding complex correspondences between ontology entities according the inten- sional meaning of their models, hence abstracting from their logical types. In this proposal, frames are considered as “unit of meaning” [79] for ontologies and are used as a mean for repres- enting intentional meaning of ontology entities. The frame-based representation of entities’ meaning enables at finding complex correspondences among entities abstracting from their logical type thus leading a step ahead the state of the art of ontology matching. Other potential benefits of this method related to understanding and generating natural language will be discussed in Section 6.3. • This thesis introduces a component-based architecture relying on Semantic Web standards for supporting knowledge-intensive tasks performed by social robots, and whose design has been guided by requirements coming from a real socially assistive robotic application. The ultimate goal aim of the architecture is to create a platform for easing the development of robotic applications by providing developers with off-the-shelf software artifacts, models and data. The strategy to pursue this goal is to massively reuse Semantic Web technologies due to their intrinsic availability and interoperability. Moreover, we present
- 29. 1.3. CASE STUDY: COMPANION ROBOTS 9 a prototype which is aimed at demonstrating feasibility and benefits of such a architecture and two applications running on top of the architecture prototype. We claim that the prototype is only an example of robotic systems that benefit of framework proposed in the thesis. This framework could potentially be integrated, with appropriate adaptions, with every autonomous agents (not limited to embodied systems). 1.3. Case Study: Companion Robots in Socially Assistive Context A case study for this thesis work has been provided by the H2020 European Project MARIO2 . This project has investigated the use of autonomous companion robots as cognitive stimulation tools for people with dementia. The MARIO robot and its capabilities were specifically designed to provide support to people with dementia, their caregivers and related healthcare professionals. Among its capabil- ities, MARIO can help caregivers in the patient assessment process by autonomously performing Comprehensive Geriatric Assessment (CGA) evaluations, and is able to deliver reminiscence therapy through personalized interactive sessions. These capabilities are part of a ro- botic software framework (inspired to the architecture presented in Chapter 7) for companion robots, and, they are supported by the knowledge representation and management framework proposed in this thesis. The overall framework and the applications presented in these thesis have been deployed on Kompaı̈-2 robots (showed in Figure 1.1), evaluated and validated during supervised trials in differ- ent dementia care environments, including a nursing home (Galway, Ireland), community groups (Stockport, UK) and a geriatric unit in hospital settings (San Giovanni Rotondo, Italy). There was a clear mutual benefit between this thesis work and the work carried out within the context of the MARIO Project. On the one hand, the MARIO benefited of the framework proposed in this thesis. On the other hand, the MARIO project provided a real-world application that fine-tuned the requirements and tested the capabilities of the contributions of this thesis. 2 MARIO project, http://www.mario-project.eu/portal/
- 30. 10 CHAPTER 1. INTRODUCTION Figure 1.1: The Kompaı̈-2 robot and its user interface. 1.4. Research Methodology The structuring and organization of the research activities have been following two complementary paths, though interlinked and interleaved among each other. On the one hand, we approached a case study with an explorative strategy aimed at investigating the needs of a real socially assistive robotic application and highlighting the limits of current solutions. On the other hand, problems that came from the real setting have been generalized in order to contribute with their solutions to advance the state of the art. The two activity paths are summarized hereafter together with the strategy for evaluating the contributions of the thesis. Explorative Approach to the Case Study. The work carried out along this path has been focusing on identifying the needs of a real socially assistive robotic application, highlighting the limits of current solutions, and, designing, developing, deploying and testing working solutions within a concrete robotic application. In line with the overall principles and methodology adopted in the project, we have been following an incremental and iterative design and development approach, inspired by Agile principles. As a consequence, the implemented approaches and solutions have been: (i) designed following a requirements-driven
- 31. 1.5. THESIS OUTLINE 11 and user-centered approach, taking into account pilot sites’ needs and scenarios; (ii) incrementally integrated, tested and validated during trial activities; (iii) gradually refined and improved on the basis of trials feedback. Research Activities and Solutions Targeting Open Problems. The work carried out along this path has been focusing on research activities aimed at the identification of solutions targeting open problems in the broad field of knowledge representation and engineering. These research problems are either inspired by and abstracted from concrete use cases, or derive from general challenges that can be specialized in the context of socially assistive robots. These problems have been synthesized in the research questions RQ0-RQ5 outlined in Section 1.1. When such a problem is identified an analysis of the current solutions is performed, and, if limitations emerge from the state of the art, then, new hypotheses are defined and tested in the real scenario. In particular, the prototype developed within the context of the MARIO project is to be interpreted as a proof-of-concept implementing the framework proposed in this thesis and evaluated in a real assistive context. Evaluation strategy. The contributions of the thesis have been eval- uated by following two different strategies, one targeting the whole robotic system and the other focusing on individual components of the architecture. A prototype of the framework presented in this thesis has been developed within the context of the MARIO project (cf. Section 1.3) and assessed during supervised trials in different dementia care environments. The evaluation of the prototype consisted of a quantitative assessment, involving the use of standardized question- naires, and a qualitative assessment, aimed at capturing impressions of the stakeholders. Architectural components were individually eval- uated through suitable experiments (meant to assess the accuracy of components) or proof-of-concepts (intended to demonstrate the feasibility of components). 1.5. Thesis Outline The remainder of this thesis is structured as follows:
- 32. 12 CHAPTER 1. INTRODUCTION Chapter 2 - Background. This chapter overviews the research areas related to this work, including a quick introduction to the Semantic Web, Socially (Assistive) Robotics and Common Sense Knowledge. Mentions of related work are also featured in other chapters. Chapter 3 - An Ontology Network for Social Robots in As- sistive Context. Chapter 3 presents a set of interconnected and modularised ontologies, called MARIO Ontology Network (MON), which are meant to model all knowledge areas that are relevant for robots’ activities in socially assistive contexts. This ontology network defines reference models for representing and structuring the knowledge processed by the robot. MON provides a robot with the means for creating, organizing, query- ing and reasoning over a background knowledge. Chapter 4 - Providing Linked Open Data as Background Knowledge for Social Robots. This Chapter investigates the possibility of populating the extensional level with data retrieved from the web. To this end two lines of research have been carried out in parallel focusing on linguistic and common sense knowledge respectively. Regarding the first line of research the chapter presents Framester, a huge linguistic knowledge graph integrating lexical, linguistic, ontological and encyclopedic data. This Chapter also introduces a novel empirical method for assessing foundational distinctions over Linked Open Data entities from a common sense perspective (e.g. deciding if an entity inherently represents a class or an instance from a common sense perspective). This method realizes the first step of a more general procedure meant to automatically generate common sense knowledge from Linked Open Data. Chapter 5 - Accessing Background Knowledge using Liz- ard. Chapter 5 presents Lizard, an Object-RDF mapper provid- ing software components with the access to the knowledge base following the Object-Oriented paradigm. Chapter 6 - A Frame-based Approach for Integrating On- tologies. Chapter 6 describes the frame-based approach for integrating ontologies. This method enables to integrate know-
- 33. 1.5. THESIS OUTLINE 13 ledge from different structured (namely, ontologies and know- ledge graphs) and unstructured sources (e.g. text). Chapter 7 - A Knowledge Base Centered Software Archi- tecture for Social Robots. This chapter presents a com- ponent-based architecture relying on semantic web technologies for supporting knowledge-intensive tasks performed by social robots. Moreover, this chapter presents a prototype which is aimed at demonstrating feasibility and benefits of such a archi- tecture and two applications running on top of the architecture prototype. Chapter 8 - Conclusion and Future Work. A summary of the overall research activity and possible lines of future work to be followed from the current state concludes this dissertation.
- 35. Chapter 2 Background 2.1. Social Robots Social Robots are embodied agents designed to socially interact with people and can be categorized depending on the application domain or depending on the type of tasks are designed to perform (for compre- hensive overviews please refer to [29, 67, 112, 171]). Leite et al. [112] identified four different application domains: health care, education, work environments and public spaces, and home. Socially Assistive Robots can be broadly classified into three categories (a review of the assistive social robots is provided by [29, 171]): (i) Service Robots are devices designed to support people living independently by assist- ing them with mobility, completing household tasks, and monitoring health and safety; (ii) Companion Robots [47] are meant to create companionship for human beings; (iii) Coaching Robots (e.g. [60, 61]) that act as a coach to encourage human beings through a series of therapeutic tasks for enhancing their health conditions. 2.1.1. Software Architectures for Social Robots Despite the different application domains and the intended functions, most of the architectures of social robots [34, 60, 61, 84, 85, 91, 95, 125, 133, 201] are constituted by the following elements: 1. A subsystem that manages the hardware devices allowing the robot to perceive the environment (such as lasers used for nav- igation, cameras, touch sensors, microphones etc.). 15
- 36. 16 CHAPTER 2. BACKGROUND 2. A set of components dealing with the robot’s motors and actu- ators (e.g. wheel engines, speakers). 3. A knowledge base storing information for supporting the ro- bot’s behaviors, tracing the users’ activities or preferences, and collecting from the operating environment (e.g. maps). 4. A multi-modal user interface that provides users with multiple modes to interact with robots. This is typically delivered by a voice-user interface and/or a touch-screen device. 5. A behavior controller that gathers information from perceptual components, knowledge base and user interfaces, and decides the next actions to perform. 6. If necessary, a supervision interface that enables to remotely control the robot and to possibly interrupt the robot’s operation. Using such a interface is part of one of the most common Human- Robot Interaction experimental techniques called Wizard of Oz [100, 168]. The architecture presented in Chapter 7 follows the structure of exist- ing architectures and defines a subsystem that allows to dynamically (i.e. at run-time without need of re-deploy) extend the capabilities of the robot, thus enabling an agile and evolutionary development of the architecture. Examples of such flexibility mechanisms can be also found in literature. Fritsch et al. [69] proposed a flexible infrastructure to extend the capabilities of the companion enabling the interaction with humans. XML is used In this proposal as language for defining the format of messages exchanged by the components and to define the sequence of operations the robot have to perform. Similarly, the extensible architecture introduced by Rossi et al. [173] allows to modify and expand the multi-modal interface without impacting the rest of the architecture. These mechanisms partially fulfill the flexibility and extensibility requirement since does not allow to dynamically deploy new software components. This feature is provided by our architecture which guarantees extensibility of the robot behaviors and capabilities. Other novel elements of our architecture concern with the use of semantic web and linked data for managing and bootstrapping the knowledge base.
- 37. 2.2. THE SEMANTIC WEB 17 Figure 2.1: The Semantic Web stack. 2.2. The Semantic Web The Semantic Web [18] is an extension of the Web aims at providing a common framework that allows data to be shared and reused across application boundaries. Standardisation for Semantic Web is under the care of World Wide Web Consortium (W3C). The W3C standards for the Semantic Web mainly include: XML, RDF(S), OWL and SPARQL. Figure 2.1 shows the semantic web stack and provides an overview of the standard technologies recommended by the W3C. 2.2.1. Extensible Markup Language (XML) Extensible Markup Language (XML) is a markup language that defines a set of rules for encoding documents in a both human-readable and machine-readable format. An XML document consists of a properly nested set of open and close tags, where each tag can have a number of attribute-value pairs. Crucial to XML is that the vocabulary of the tags and their allowed combinations is not fixed, but can be defined per application of XML. In the Semantic Web context, XML is being used as a uniform data-exchange format thus providing a common syntax for exchange data across the web.
- 38. 18 CHAPTER 2. BACKGROUND 2.2.2. Resource Description Framework (RDF) Resource Description Framework (RDF)1 is a W3C recommendation originally designed as metadata model, it has being used as a general framework for modelling information. The basic construction in RDF is the triple <subject, preficate, object>. The subject denotes a resource and the predicate expresses a relationship between the subject and the object (which can be a value or another resource). For example, a way for representing the fact “The author of War and Peace is Leo Tolstoy” is :War and Peace :author :Leo Tolstoy where :War and Peace and :Leo Tolstoy are the Uniform Resource Identifiers (URIs) of two resources representing respectively the book titled “War and Peace” and the writer “Leo Tolstoy”, and :author is the URI of the predicate “author” which is used to connect a book to its author. It is easy to see that an RDF model can be seen as a graph where nodes are values or resources and edges are properties. Several common serialisation formats of RDF are in use, including: TURTLE2 , RDF/XML3 , N-Triples4 . RDF Schema (RDFS)5 provides a data-modelling vocabulary for RDF data. RDFS is an extension of RDF aims at providing basic elements for structuring RDF resources. It allows to define: Classes, Properties, Datatypes and Hierarchies for both classes and properties. 2.2.3. Web Ontology Language (OWL) The Web Ontology Language (OWL)6 is a semantic markup language for defining, publishing and sharing ontologies on the World Wide Web. OWL can be used to explicitly represent the meaning of terms in vocabularies and the relationships between those terms. This representation of terms and their interrelationships is called ontology. OWL is part of the Semantic Web stack (see Figure 2.1) and it is complementary to XML, RDF and RDFS: 1 RDF, W3C Recommendation https://www.w3.org/TR/rdf11-concepts/ 2 TURTLE, https://www.w3.org/TR/turtle/ 3 RDF/XML, https://www.w3.org/TR/rdf-syntax-grammar/ 4 N-Triples, https://www.w3.org/TR/n-triples/ 5 RDFs, W3C Recommendation https://www.w3.org/TR/rdf-schema/ 6 OWL, W3C Recommendation https://www.w3.org/TR/owl-ref/
- 39. 2.2. THE SEMANTIC WEB 19 • XML provides a surface syntax for structured documents, but imposes no semantic constraints on the meaning of these docu- ments; • RDF is a datamodel for resources and relations between them. It provides a simple semantics for this datamodel; • RDFS is a vocabulary for describing properties and classes of RDF resources, with a semantics for generalisation-hierarchies of such properties and classes; • OWL adds constructs for describing properties and classes: among others, relations between classes (e.g. disjointness), car- dinality (e.g. “exactly one”), equality, richer typing of properties, characteristics of properties (e.g. symmetry), and enumerated classes. 2.2.4. SPARQL SPARQL7 is a query language for retrieving and manipulating data store in RDF format. Most forms of SPARQL queries contain a set of triple patterns called “basic graph pattern”. Triple patterns are like RDF triples except that each of the subject, predicate and object may be a variable (denoted by a question mark). A basic graph pattern matches a subgraph of the RDF data when RDF terms from that subgraph can be substituted with the variables of the pattern. For example, the following SPARQL query retrieves pairs books authored by Tolstoy. SELECT ?book WHERE {?book :author :Leo Tolstoy} 2.3. Ontologies Historically ontology, listed as part of metaphysics, is the philosophical study of the nature of being, becoming, existence, or reality, as well as the basic categories of being and their relations. Ontology deals with questions concerning what entities exist or can be said to exist, and how such entities can be grouped, related within a hierarchy, 7 SPARQL, W3C Recommendation https://www.w3.org/TR/ rdf-sparql-query/
- 40. 20 CHAPTER 2. BACKGROUND and subdivided according to similarities and differences. While the term ontology has been rather confined to the philosophical sphere in the recent past, it has gained a specific role in a variety of fields of Computer Science, such as Artificial Intelligence, Computational Linguistics, and Database Theory and Semantic Web. In Computer Science the term loses part of its metaphysical background and, still keeping a general expectation that the features of the model in an ontology should closely resemble the real world, it is referred as a formal model consisting of a set of types, properties, and relationship types aimed at modeling objects in a certain domain or in the world. In early ’90s Gruber [87] gave an initial and widely accepted definition: An ontology is a formal, explicit specification of a shared conceptual- ization. An ontology is a description (like a formal specification of a program) of the concepts and relationships that can formally exist for an agent or a community of agents. Accordingly, ontologies are used to encode a description of some world (actual, possible, counterfactual, impossible, desired, etc.), for some specific purpose. In the Semantic Web, ontologies have been used as a formalism to define the logical backbone of the Web itself. The language used for designing ontologies in the Web of Data is the Web Ontology Language (OWL). In the last decade there has been a lot of research for investigating best practices for ontology design and re-use in the Web of Data. Among the others the EU-FP7 NeOn project8 has provided sound principles and guidelines for designing complex knowledge networks called ontology networks. An ontology network is a set of interconnected ontologies. According to [4], the interconnections can be defined in a variety of ways, such as alignments, modularization based on owl:import axioms, and versioning. Ontology networks enable modular ontology design in which each module conceptualizes a specific domain and can be designed by using Ontology Design Patterns [78] and pattern-based ontology design methodologies, such as eXtreme Desing [23]. 8 http://www.neon-project.org/
- 41. 2.3. ONTOLOGIES 21 2.3.1. Knowledge Management Frameworks for Social Robots Ontologies and Semantic Web technologies can support the develop- ment of robotic systems and applications that deal with knowledge representation, acquisition and reasoning. Furthermore, Semantic Web standards enable the interlinking of local robotic knowledge with available information and resources coming from the Web of Data. This trend has also led to the creation of the IEEE RAS Ontologies for Robotics and Automation Working Group (ORA WG), with the goal of developing a core ontology and an associated methodology for knowledge representation and reasoning in robotics and automa- tion [161]. In this direction, different frameworks have been proposed to model, manage and make available heterogeneous knowledge for robotic sys- tems and applications. Focusing on service robots that operate in indoor environments through perception, planning and action, the ontology-based unified robot knowledge framework (OUR-K) [117] aims at supporting robot intelligence and inference methods by in- tegrating low-level perceptual and behavioural data with high-level knowledge concerning objects, semantic maps, tasks, and contexts. An ontology-based approach is also adopted in the ORO knowledge man- agement platform [113]. The platform stores and processes knowledge represented according to the OpenRobots Common Sense Ontology9 , an OWL ontology based on the OpenCyc upper ontology and extended with the definition of reference concepts for human-robot interaction. When deployed on a robot, the knowledge base can be instantiated with a priori common-sense knowledge and is then used as a “semantic blackboard” where the robotic modules (such as the perception module, the language processing module, the task planner and the execution controller) can store the knowledge they produce and query it back. Along the same path, research projects and initiatives, such as KnowRob10 , RoboEarth11 and RoboBrain12 , go beyond local know- ledge bases and, also with the emergence of cloud-based robotics, pro- pose Web-scale approaches. KnowRob [195, 196, 197] is a knowledge 9 https://www.openrobots.org/wiki/oro-ontology 10 http://knowrob.org/ 11 http://roboearth.ethz.ch/ 12 http://robobrain.me/
- 42. 22 CHAPTER 2. BACKGROUND processing system and semantic framework for integrating information from different sources, including encyclopedic knowledge, common- sense knowledge, robot capabilities, task descriptions, environment models, and object descriptions. Knowledge is represented and form- ally modeled according to a reference upper ontology, defined using the Web Ontology Language (OWL). The system supports different reas- oning capabilities and provides interfaces for accessing and querying the KnowRob ontology and knowledge base. Similarly, the RoboEarth framework [198] provides a web-based knowledge base for robots to access and share semantic representations of actions, object models and environments, augmented with rule-based learning and reasoning capabilities. The RoboEarth knowledge base relies on a reference ontology, as an extension of the KnowRob ontology to (i) represent actions and relate them in a temporal hierarchy; (ii) describe object models to support recognition and articulation; and (iii) represent map-based environments. An HTTP-based API enables robots to access the knowledge base for uploading, searching and downloading information from and to their local knowledge bases. Along the same path, the Robo- Brain knowledge engine [176] aims at learning and sharing knowledge gathered from different sources and existing knowledge bases, includ- ing linguistic resources, such as WordNet, image databases, such as ImageNet, and Wikipedia. Although the RoboBrain knowledge base does not explicitly adopt ontologies and Semantic Web technologies, knowledge is represented in a graph structure and stored in a graph database. A REST API enables robots to access RoboBrain as-a- service, to provide and retrieve knowledge on the basis of a specific query language. The need to provide robots with a knowledge representation and management framework able to handle knowledge from different sources (including external data sources and knowledge bases) and support multiple tasks and applications has long been considered in robotics. However, it is only in recent years that the potential of ontology-based knowledge representation approaches and Semantic Web technologies has been considered to address the two aforemen- tioned points in robotic platforms.
- 43. 2.4. PATTERN-BASED ONTOLOGY DESIGN 23 2.4. Pattern-based Ontology Design The notion of “pattern” has proved useful in design, as exemplified in diverse areas, such as software engineering. Under the assumption that there exist classes of problems that can be solved by applying common solutions (as has been experienced in software engineering), it is suggested to support reusability on the design side specifically. To this end Ontology Design Patterns (ODPs) have been proposed as modeling solutions to recurrent ontology design problems. ODPs are modeling components that can be used as basic building blocks of an ontology network. eXtreme Design (XD) is an ontology design methodology that supports the pattern-based approach. We adopted XD as methodology for designing the MARIO Ontology Network presented in Chapter 3 and we extensively reused ODPs. Sections 2.4.1 and 2.4.2 briefly introduce ODPs and XD, respectively. 2.4.1. Ontology Design Patterns Ontology Design Patterns (ODPs) [78] is an emerging technology that favors the reuse of encoded experiences and good practices. ODPs are modeling solutions to solve recurrent ontology design problems. They can be of different types including: (i) logical, which typically provide solutions for solving problems of expressivity e.g., expressing n-ary relations in OWL; (ii) architectural, which describe the overall shape of the ontology (either internal or external) that is convenient with respect to a specific ontology-based task or application e.g. a certain DL family; (iii) content, which are small ontologies that address a specific modeling issue, and can be directly reused by importing them in the ontology under development e.g., representing roles that people can play during certain time periods; (iv) presentation, which provide good practices for e.g. naming conventions. 2.4.2. eXtreme Design eXtreme Design (XD) [163, 23, 164] is a family of methods and associated tools, based on the application, exploitation, and definition of ontology design patterns (ODPs) for solving ontology development issues. XD principles are inspired by those of the agile software methodology called eXtreme Programming (XP). The main idea of agile software development is to be able to incorporate changes easily,
- 44. 24 CHAPTER 2. BACKGROUND Figure 2.2: An example of collected uses-story. in any stage of the development. Instead of using a waterfall-like method, where you first do all the analysis, then the design, the implementation and finally the testing, the idea is to cut this process into small pieces, each containing all those elements but only for a very small subset of the problem. XD is test-driven, and applies the divide-and-conquer approach as well as XP does. Also, XD adopts pair design, as opposed to pair programming. The main principles of the XD method can be summarised as follows: • Customer involvement and feedback. The customer should be involved in the ontology development and its representative should be aware of all parts of the ontology project under devel- opment. Interaction with the customer representative is key for favoring the explicit expression of the domain knowledge. • Customer stories and Competency Questions. The onto- logy requirements and its tasks are described in terms of small stories by the customer representative. Designers work on those small stories and, together with the customer, transform them in the form of Competency Questions [88] (CQs). CQs will be used through the whole development, and their definition is a key phase as the designers have the challenge to help the customer in making explicit as much implicit knowledge as possible. We asked all the partners involved in the case study to provide their own stories. The template for providing the stories is shown
- 45. 2.4. PATTERN-BASED ONTOLOGY DESIGN 25 in Figure 2.2. The fields “Partner”, “Scriber”, “e-mail” were used for asking further clarification about the story. The Title field helped for a better understanding the main focus of the story. The “Priority” field was used to choose the stories to treat first. The allowed values were High, Medium and Low. “Depends on” allowed to specify a link between two stories. For example, if a story was too long, it could be split into two stories and this field allowed one to express the dependency. The last field “Knowledge area(s)” was used for associating the story with one or more knowledge areas which the story belonged to. The customer stories collected together with the resulting Com- petency Questions can be retrieved on-line13 . Other competency questions have been extracted by analysing domain documents, such as those used for effectuating a Comprehensive Geriatric Assessment (CGA) of a patient. • Content Pattern (CP) reuse and modular design. A de- velopment project is characterised by two main sets: (i) the problem space composed of the actual modelling issues that have to be addressed during the project which are called “Local Use Case” (LUC); (ii) the solution space made up of reusable model- ling solutions, called “Global Use Case” (GUC), representing the problem that a certain ODP provides a solution for. If there is a CP’s GUC that matches a LUC it has to be reused, otherwise a new module is created. An analysis of the possible strategies for reusing CP is provided by [164]. • Collaboration and Integration. Collaboration and constant sharing of knowledge is needed in a XD setting, in fact similar or even the same CQs and sentences can be defined for different stories. When this happens, it means that these stories can be modelled by reusing a set of shared CPs. • Task-oriented design. The focus of the design is on that part of the domain of knowledge under investigation that is needed in order to address the user stories, and more generally, the tasks that the ontology is expected to address. • Test-driven design. A new story can be treated only when 13 http://etna.istc.cnr.it/mario/D5.1/.
- 46. 26 CHAPTER 2. BACKGROUND all unit tests associated with it have been passed. An ontology module developed for addressing a certain user story associated to a certain competency question, is tested e.g. (i) by encoding in the ontology a sample set of facts based on the user story, (ii) defining one or a set of SPARQL queries that formally encode the competency question, (iii) associating each SPARQL query with the expected result, and (i) running the SPARQL queries against the ontology and compare actual with expected results. 2.5. Ontology Matching Among the various semantic technology proposed to handle heterogen- eity Ontology Matching [181] has proved to be an effective solution to automate integration of distributed information sources. Ontology Matching (OM) finds correspondences between semantically related entities of ontologies. These correspondences enable several tasks such as ontology merging, query answering, or data translation. There have been proposed several formalization of the matching problem, we follow the formalization in [57] that provide a unified approach over the previous works. The matching problem is the problem of finding an alignment between two ontologies. An alignment is a set of 4-uple (e1, e2, r, n) where: (i) e1 and e2 are entities defined by the first and the second ontology, respectively; (ii) r is a relation holding between e1 and e2, e.g., equivalence, subsumption, disjointness; (iii) n is the confidence measuring the likelihood that the relation holds. 2.6. Linguistic Linked Open Data Resources Many resources belonging to different domains are now being pub- lished on-line using Linked Data principles to provide easy access to structured data on web. This includes many linguistic resources that are already a part of Linked Data, but made available mainly for the purpose of being used by NLP applications. Two of the most important linguistic linked open data resources are WordNet [135] and FrameNet [10]. They have already been form- alised as semantic web resources, e.g. in OntoWordNet [75], WordNet RDF [7], FrameNet RDF [143], etc. FrameNet allows to represent textual resources in terms of Frame Semantics. The usefulness of
- 47. 2.7. COMMON SENSE KNOWLEDGE 27 FrameNet is limited by its limited coverage, and non-standard se- mantics. An evident solution would be to establish valid links between FrameNet and other lexical resources such as WordNet , VerbNet and BabelNet to create wide-coverage and multi-lingual extensions of FrameNet. By overcoming these limitations NLP-based applications such as question answering, machine reading and understanding, etc. would eventually be improved. Within MARIO these were important requirements, hence we developed Framester (presented in Chapter 3): a frame-based ontological resource acting as a hub between e.g. Fra- meNet, WordNet, VerbNet, BabelNet, DBpedia, Yago, DOLCE-Zero, and leveraging this wealth of links to create an interoperable predicate space formalised according to frame semantics [65], and semiotics [70]. Data designed according to the predicates in the predicate space created by Framester result to be more accessible and interoperable, modulo alignments between specific entities or facts. The closest resources to Framester are FrameBase [174] and Pre- dicate Matrix [106]. FrameBase aimed at aligning linked data to FrameNet frames, based on similar assumptions as Framester’s: full- fledged formal semantics for frames, detour-based extension for frame coverage, and rule-based lenses over linked data. However, the cov- erage of FrameBase is limited to an automatically learnt extension (with resulting inaccuracies) of FrameNet-WordNet mappings, and the alignment to linked data schemas is performed manually. Anyway, Framester could be combined with FrameBase (de)reification rules so that the two projects can mutually benefit from their results. Predicate Matrix is an alignment between predicates existing in FrameNet, VerbNet, WordNet, and PropBank. It does not assume a formal semantics, and its coverage is limited to a subset of lexical senses from those resources. A RDF version of Predicate Matrix has been created in order to add it to the Framester linked data cloud, and (ongoing work) to check if those equivalences can be reused in semantic web applications. 2.7. Common Sense Knowledge Over the years, a number of projects aimed at generating common sense knowledge. Regardless specific settings, the results of these projects can be seen an ontology O =< T, A > consisting of T, a T-box (i.e. terminology box, also called schema or vocabulary) and A
- 48. 28 CHAPTER 2. BACKGROUND is an A-box (assertion box). The outcome of these projects can be informally classified on the basis of following criteria. (i) the process used to generate knowledge (e.g. automatic or with humans in the loop); (ii) the breadth of the knowledge produced (i.e. universal or domain specific); (iii) the formalism used to represent knowledge; (iv) the density of the knowledge (i.e. the number of assertion per entity); (v) the richness (the variety of types and relations) and the depth (the number of inheritance relations) of the t-box; (vi) the level of interoperability with other datasets (i.e. the linkage of the ontology at both intensional and extensional level); (vii) the metadata (i.e. provenance and validity of stated facts). Among existing projects we overview strengths, weaknesses and results of DBpedia, ConceptNet, NELL. DBpedia14 [21] is a very popular dataset automatically obtained from Wikipedia infoboxes. DBpedia is the de-facto main hub of the Web of Data containing 4,58 milion entities of encyclopedic nature. However, some weaknesses of DBpedia are the scarcity of relations among entities and the limited depth of the vocabulary. In fact, the DBPedia ontology is induced from Wikipedia and is only partially aligned with existing formal theories (e.g. foundational ontologies). These drawbacks make hard answer queries such that “what knives are used for?”, “give me all physical entities”. ConceptNet15 [120, 187, 186] is a large scale multilingual semantic network that integrates knowledge from (i) Open Mind Common Sense project [183, 184] who ran a web site that collected common sense facts from users; (ii) existing datasets such as DBpedia, Wiktionary, Open Multilingual WordNet, OpenCyc [115] and Umbel; (iii) Verbosity, a game with a purpose that learns common sense knowledge from people’s intuitive word associations. ConceptNet defines thirty lexical and common sense relations among its entities such as: “antonym”, “synonym”, “is used for” (that associates an object with what is used for, e.g. “bridge” and “cross water”), “at location” (that associates an object with its typical locations, “butter” and “refrigerator”), “capable of” (associating an object with what it can do, e.g. “knife” and “cut”) etc. NELL16 [35, 138] is an ongoing project aiming at learning large semantic network (similar to ConceptNet) with a never 14 DBpedia, http://wiki.dbpedia.org/ 15 ConceptNet, http://conceptnet.io/ 16 NELL, http://rtw.ml.cmu.edu/rtw/
- 49. 2.7. COMMON SENSE KNOWLEDGE 29 ending approach. The main drawback of ConceptNet and NELL is the inherent ambiguity of their concepts (e.g. “apple” is both a fruit and a computer company that can be used both for “eating” and for “computing”, and it is controlled by “Steve Jobs”). ConceptNet and NELL provide information about provenance and confidence of facts, but they miss contextual conditions that make the facts true. Moreover, it is not clear how these projects select the resources to extract the knowledge from.
- 51. Chapter 3 An Ontology Network for Social Robots in Assistive Context In order to interact with people showing human-like features, a social robot must be provided with a human-like background knowledge. Furthermore, when employed in socially assistive context, robots con- tinuously perform knowledge-intensive tasks aimed at (i) assisting their users with their daily activities (e.g. drive the patients to a specific location or identifying searched objects); (ii) helping nurses, physician and familiars in the healthcare process of people with de- mentia (e.g. collecting information for assessing patient’s cognitive status). Determining what kind of knowledge social robots need in socially assistive context and how to organize their knowledge is the goal related to research question RQ1 (cf. Section 1.1) which is in- vestigated in this chapter. To this end, the chapter presents a set of interconnected and modularised ontologies, i.e. the MARIO Ontology Network (MON), which are meant to model all knowledge areas that are relevant for robots’ activities in socially assistive contexts. This ontology network defines reference models for representing and struc- turing the knowledge processed by the robot. MON provides a robot with the means for creating, organizing, querying and reasoning over a background knowledge. The robot background knowledge consists of: lexical knowledge (e.g. natural language lexica and linguistic frames), domain knowledge (e.g. users related information), environmental knowledge (e.g. physical locations and maps), sensor knowledge (e.g. RFID, life measures), and metadata knowledge (e.g. entity tagging). The Ontology Network, named MARIO Ontology Network (MON), is composed of several modularised ontologies that cover different know- 31
- 52. 32 CHAPTER 3. AN ONTOLOGY NETWORK FOR SOCIAL ROBOTS ledge areas that are relevant to the tasks of supporting people affected by dementia. The knowledge areas and the ontology modules were identified by analysing the use cases that emerged from the MARIO project (cf. Section 1.3), These uses cases mainly describe actions and behaviors featuring the MARIO robot. Nevertheless, they also provide a detailed descriptions about the nature of the knowledge that the robot has to deal with in order to behave. The MON consists of 53 modules covering 12 knowledge areas. The Ontology Network has been developed following the eXtreme Design methodology (introduced in Section 2.4.2) and by extensively reusing Ontology Design Patterns (cf. Section 2.4.1). The rest of the Section is organized as follows. Section 3.1 describes the ontology development process, Section 3.2 presents the MON’s knowledge areas and Section 3.3 outlines the most innovative ontology modules of the Ontology Network. 3.1. Design Methodology The MARIO Ontology Network (MON) has been designed by following best practices and pattern-based ontology engineering methods aimed at extensively re-using Ontology Design Patterns (ODPs) [78]. In particular, the MON has been developed following the eXtreme Design (XD) [23, 163] methodology (cf. Section 2.4.2). This methodology has been extended in order to identify the knowledge areas that are relevant to a companion robot (cf. 3.2) and to provide ontology engineers with guidelines for re-using existing ontologies (cf. 3.1.1). Section 3.1.2 illustrates how eXtreme Design has been configured in for the development of MON. 3.1.1. Guidelines for Ontology Re-use Linked Data is rapidly increasing, especially in the public sector where opening data is becoming a consolidated institutional activity. However, the importance of providing Linked Data with a high quality ontology modeling is still far from being fully perceived. The result is that Linked Data are mostly modeled by directly reusing individual classes and properties defined in external ontologies, overlooking the possible risks caused by such a practice. Although ontology reuse is a recommended practice in most ontology design methodologies [182], a standardization of ontology reuse practices is still missing. Most
- 53. 3.1. DESIGN METHODOLOGY 33 Figure 3.1: The eXtreme Design workflow as extended in [164]. The highlighted tasks involve the guidelines for ontology reuse. literature on ontology reuse is focused on the challenging issue of ontology selection, while our perspective is on how to implement reuse once the selection finalized. This practice may compromise the level of semantic interoperability that can be achieved. Therefore, the need of clear guidelines for ontology reuse arise. In [164] we provided a series of guidelines for ontology reuse in the context of ontology projects that exhibit these characteristics: (i) there is no ontology that addresses all or most of the requirements of the local ontology project; (ii) the ontology under development is meant to be used as a reference ontology for a certain domain; (iii) there is the willingness to comply with existing standards. These guidelines can be integrated into the tasks 7 and 8 of the XD workflow (cf. Figure 3.1 and Section 2.4.2). Ontology re-use models can be classified based on (i) the type of reused ontology (e.g. foundational, top-level, ontology design pat- terns, domain ontologies); (ii) the type of reused ontology fragment (e.g. individual entities, modules, ontology design patterns, arbitrary fragments); (iii) the amount of reused axioms (e.g. import of all axioms, of only axioms in a given neighbourhood of an entity, of no ax- ioms); (iv) the alignment policy (e.g. direct reuse of entities, reuse via
- 54. 34 CHAPTER 3. AN ONTOLOGY NETWORK FOR SOCIAL ROBOTS equivalence or subsumption relations such as owl:equivalentClass and rdfs:subClassOf). The only characteristic that all these models share is to reuse entities with the same logical type as they were defined (e.g. an entity defined as owl:Class in an ontology is commonly reused as such). We identify the following possible approaches to ontology reuse. Direct Reuse of Individual Entities. This approach consists of directly introducing individual entities of external ontologies in local axioms. This practice is very common in the Linked Data community, however it is a routine, not a good practice, at all. It is essentially driven by the intuition of the semantics of concepts based on their names, instead of their axioms. In this case, the risk that the formal semantics of the reused entities is incompatible with the intended semantics to be represented is rather high. Moreover, with this practice a strong dependency of the local ontology with all the reused ontologies is created. This dependency may put at risk the sustainability and stability of the local ontology and its associated knowledge bases: if a change in the external ontology introduces incoherences in the local one, they must be dealt with a redesign process and consequential change in the ontology signature. Indirect Reuse of Ontology Modules and Alignments. With this ap- proach, the modeling of some concepts and relations, which are relevant for the domain but applicable to more general scopes, is delegated to external ontologies by means of ontology module reuse. An ontology module is a fragment that may be identified as providing a solution to one or more specific requirements of the local ontology. For ex- ample, let us consider an external ontology modeling the participation of an individual (e.g. through a property ex:isInvolvedIn) to an event (e.g. a class ex:Event). If the local ontology needs to spe- cify a particular involvement in an event (e.g. lo:hosted) it should specialize (it indirectly reuses) the relation of the external one (i.e. ex:isInvolvedIn). The fragment of the external ontology identified as relevant for the local ontology may be communicated in some usage documentation provided with the ontology. Nevertheless, it is difficult to provide third parties with a formal indication of the fragment that was meant to be relevant. This may lead to high heterogeneity in the usage of external fragments in data modeled through the local ontology. As for ontology sustainability, when a change in the external
- 55. Another Random Scribd Document with Unrelated Content
- 56. remembered that these families were those who have worked their way through the first difficulties, these figures become doubly significant. There is, for example, a Ukrainian family from the Russian Ukraine. It consists of the parents and four children between the ages of three and fifteen. Ever since the family came to the United States they have had one or more lodgers to help them pay the rent. At present they have three men paying four dollars a month each; and as the father, who had been working in the stockyards for nineteen dollars a week, was discharged two months ago, the wife has been working in a spring factory to support the family. Then there is a Polish family, composed of the parents and four children under fourteen, two of them children of the man by a former wife. The father has been in this country since 1894, but his wife has been here only since 1910. For two years after their marriage the wife worked at night, scrubbing from 6.30 to 9.30 p.m., and received twenty-four dollars a month. Then there was an interval while her children were babies, during which she did not work, but the family lived on the earnings of the father. For the last two years, however, his work has been slack, first because of a strike, and later due to an industrial depression in his trade, and the mother is again at work, this time in a tailor shop, earning ten dollars a week. The effect of the mother's work in decreasing the child's chances for life has been made clear by the studies of the Children's Bureau in Johnstown,[10] Montclair,[11] and Manchester,[12] in all of which a higher rate of infant mortality was shown for children of mothers gainfully employed. The effect of the mother's work on the family relationship and the home life of the family group is, of course, not measurable in absolute terms. The leaders of the various national groups, however, have repeatedly emphasized the fact that the absence of the women from the home has created entirely new problems in the family life.
- 57. They have pointed out that while the peasant women have been accustomed to work in the fields in the old country, their work did not take them away from their homes as the work in this country does. If they were away there was usually some older woman to take care of the children. Here the work of the mother frequently results in neglect of the children and the home. In recognition of this fact attempts have been made to solve the problem. Among Slovenians it was customary, before the war made it impossible, to send the children back to the old country to their grandmothers to be cared for. One priest said he had seen women taking as many as twelve children to a single village. The Ukrainians in Chicago have talked of establishing a day nursery to look after their children, but the people are poor, and it has not been possible to raise the money. In the meantime children are not sent to the day nurseries already established, but are commonly taken to neighbors, some of whom are paid for taking care of ten or twelve children. This arrangement constitutes a violation of the city ordinance requiring day nurseries to be licensed, but is evidently a violation quite unconsciously committed by both parties to the transaction. A group of nonworking Lithuanian women heard that neglected children were reported to the settlement in the neighborhood. One of the women investigated, and found many children locked in houses for the day, with coffee and bread for lunch. One child, too small to shift for himself, was found with his day's supply of food tied around his neck. The women decided to open a nursery in charge of a Lithuanian woman who would be able to speak to the children in their own language, as few children below school age spoke English. The original plans were to accommodate ten or twelve children, but as soon as the nursery opened there were so many women wanting to leave their children there that it took as many as thirty children. The nursery was maintained for about eighteen months, and was then closed because of the difficulty of raising the necessary funds. Some such plan must be developed that takes care of the foreign- born mother's work if she is forced to supplement the family's
- 58. income outside of the home. The organization of family life that has grown up parallel with the industrial system assumes her presence in the home. When misfortune makes this impossible some provision for caring for the children must be found. CHANGED DUTIES OF A MOTHER Another changed condition in the life in this country is that the family group is usually what the sociologist calls the "marriage" group, as distinguished from the "familial" group, which is generally found in the old country. The grandmothers and maiden aunts, who were part of the group in the old country, and who shared with the mother all the work of the household, are not with them in this country. The older women are seldom brought on the long journey, and the maiden aunt is either employed in the factory system, or she sets up a house of her own, so that in any event her assistance in the work of the household can no longer be relied on. It is perhaps the grandmother that is missed the more, because it was to her that the mother of a family was wont to turn for advice as well as assistance. This decrease in the number of people in the household is not compensated for by the diminution in the amount of work, which is another fact of changed conditions. For in this country the housewife no longer spins and weaves, or even, as a rule, makes the cloth into clothing. She does not work in the fields, or care for the garden or the farm animals, all of which she was expected to do in the old country. The loss of the older women in the group, however, means that what tasks are left must all be done by her. The duties of the housewife may not be as many, but the work they involve may be more. This is true, for example, in the matter of feeding the family. In Lithuania soup was the fare three times daily, and there were only a few variations in kind. Here the family soon demands meat, coffee, and other things that are different from the food she has cooked in the old country.... Occasionally the situation is further complicated by the insistence of dietetic experts that the
- 59. immigrant mother cannot feed her family intelligently unless she has some knowledge of food values. In other words, the work of the housewife was easy in the old country because it was well done—if it was done in the way her mother did it—and conformed to the standards that she knew. It could thus become a matter of routine that did not involve the expenditure of nervous energy. Here, on the other hand, she must conform to standards that are constantly changing, and must learn to do things in a way her mother never dreamed of doing them. And there is the new and difficult task of planning the use of the family income, which takes on a new and unfamiliar form. In spite of all that has been taken out of the home the duties of the housewife remain manifold and various. She is responsible for the care of the house, for the selection and preparation of food, for spending the part of the income devoted to present needs, and for planning and sharing in the sacrifices thought necessary to provide against future needs. She must both bear and rear her children. The responsibilities and satisfactions of her relationship with her husband are too often last in the list of her daily preoccupations, but by no means least in importance, if one of the essentials of a home is to be maintained. The enumeration of the tasks of any wife and mother throws into relief the difficulties of the foreign-born mother. The all too frequent cases where homes are deprived of her presence emphasize how indispensable she is. All case-work agencies have had to grapple with the problem of families suffering this deprivation. It is these motherless families that make us realize how many tasks and responsibilities fall to the lot of the mother. There was a motherless Russian family, consisting of the father and six children, the oldest a girl of thirteen and the youngest a five- month-old boy. For a time the family tried to get along without asking advice of an outside agency. The baby was placed with friends, and the thirteen-year-old girl stopped school to care for the five-room flat and the other four children. In a short time the family
- 60. with whom the baby was placed wanted to adopt him, and refused to keep him longer on any other condition. At this time the Immigrant's Protective League was appealed to for help in placing the baby where he would not have to be given for adoption. They found the father making a pathetic attempt to keep the home and children clean, and the oldest girl, Marya, trying hard to take her mother's place. The best plan they were able to work out for the family was institutional care for the youngest two children, nursery care outside of school hours for the next two, and the two oldest left to take care of themselves, although given lunch at the school. Marya, of course, was sent back to school, and she and her father share the housekeeping. PATERNAL AUTHORITY PASSING A third change should be taken into account. There is a marked difference between the general position of women and children in relation to the authority of the husband and father in this country and that in the old country. It is indicated in both general opinion and express statutory amendment in this country, although not in the so-called common law. The latter, in common with practice in the native lands of immigrants, provided that marriage gave the husband the right to determine where the domicile should be, the right "reasonably to discipline" wife and children, the right to claim her services and to appropriate her earnings and those of the children, the right to take any personal property (except "paraphernalia" and "pin money") she might have in full ownership, the right to manage any land she might become entitled to, and the right to enjoy the custody of the children, regardless of the maintenance of his conjugal fidelity, in the absence of such obscene and drunken conduct on his part as would be obviously demoralizing to the young child. There existed no adequate provision for enforcing the father's performance of either conjugal or parental obligations, and the result has been the development of two bodies of legislative change. One of these has granted to the wife certain rights as against the
- 61. husband, on the theory that the wife retains her separate existence after marriage and should retain rights of individual action. The other body of statutes imposes on the man the duty of support, making abandonment or refusal to support punishable by fine or imprisonment, or both. The theory of this legislation is that the support of wife and children is to be a legally enforceable duty, which may rightly be laid upon the man because of his special interest and special ability. Moreover, through the establishment of the juvenile court, the community has undertaken, not only to say that support must be given, but to set a standard of "proper parental care" below which family groups are not to be allowed to sink and still remain independent and intact. By creating the juvenile probation staff, an official assistant parent is provided. In the same way, by authorizing commitment of children to institutions, the dissolution of the home that falls persistently to too low a standard is made possible. The common law, as accepted in the various states, was not entirely uniform, but it was substantially the universal family law; now the states differ widely in the body of statutory enactments developed in this field. All have some laws recognizing the claims of children to have their home conditions scrutinized—though they may have no express juvenile-court law, all recognize to some extent the separate existence of the married women—though only twenty-one have given the mother substantial rights as against the father over their children, and they all recognize the parent's duty to secure the child's attendance at school, and have imposed some limitation on the parent's right to set his young child to work. In other words, in all the states the idea of the separate existence of the wife and of the interest of the community in the kind of care given the child has been embodied in legislation. These statutes have been enacted by legislatures composed largely, if not exclusively, of men, and register the general change in the community attitude toward the family group. An unlimited autocracy is gradually becoming what might now be termed a constitutional
- 62. democracy. But the law of the jurisdictions from which most of the immigrant groups come, undoubtedly represents a theory of family relationship not widely different from that underlying the common law. The South Italian group, in which the right of the father to discipline wife and daughter is passed on to the son, may represent an extreme survival of the patriarchal idea; but almost all the foreign-born groups hold to the dominion of man over woman, and of parents over children. Immigrant groups evidence their realization of the changed conditions in different ways. Among the Ukrainians in Chicago, for example, it is said that, whereas in the old country the men kept complete control of the little money that came in, here they very generally turn it all over to their wives. Some of them have laughed, and said that America was the "women's country." Among other groups, notably the Jugo-Slav and the Italian, there is said to be a general attempt to keep the women repressed and in much the same position they held in the old country. Sometimes the woman perceives the difference in the situation more quickly than her husband. Then if he attempts to retain the old authorities in form and in spirit, she may submit or else she may gradually lead him to an understanding. But she may not understand and yet may rebel and carry her difficulty to the case-work agency. One of the settlements in Chicago is said to have become very unpopular with the men in its neighborhood, as it has the reputation of breaking up families, because women who have been ill treated by their husbands have gone to the settlement to complain, and have there been given help in taking their complaints to court. The Immigrant's Protective League in Chicago receives many complaints from women who have learned that their husbands have not the right to beat them or their children. One Lithuanian woman, who had been in this country six years, came to the league with the statement that her husband often threw her and their eight-year-old son out of the house in the middle of the night. Another Lithuanian woman living in one of the suburbs took her three children and
- 63. came to Chicago to her sisters, because her husband abused her, called her vile names, and beat her. When the husband was interviewed he agreed not to do so again, and his family returned to him. Of course, the theory underlying even the feminist "married woman's property laws" included not only her enjoyment of rights, but her exercise of legal responsibility; but the restrained exercise of newly acquired freedom is evidence of high social and personal development. And the women in the foreign-born groups come from the country, the village, the small town. They have had little education, their days have been filled with work, so that there has been little time for reflection, they come from a simple situation in which there was little temptation to do wrong. They find here, on the other hand, a situation which is complex in the extreme, and in which there are elements that tend to make matters especially difficult for women. Attention has already been called to the confusion created by the lodger in the home and the special temptation to the woman to desert her husband for the lodger. The relative scarcity of women in the group, the presence of large numbers of men who cannot enter a legal marriage relationship because they have wives in the old country, the spiritual separation that often results from physical separation caused by the man's coming ahead to prepare a place— all these are undoubtedly factors that enter in to make difficult the wise use of her freedom. Native endowment, moral as well as physical and mental, varies among these women as among other women. Confronted with this confused and difficult situation, the change from the old sanctions, the old safeguards, even the old legal obligations, is difficult. It is inevitable that a few will find themselves unequal to the task of readjusting their lives. The father of one family came to the Immigrant's Protective League in Chicago, asking help because his wife had turned him out of his home. He said that she drank and was immoral. Instead of caring for the home and the two-year-old
- 64. child, she spent her time behind the bar in her brother's saloon, having "a good time" with the customers. She had deserted six weeks before, but he had found her and had had her in the Court of Domestic Relations, where he had been persuaded to take her back. He said she was still drinking and still neglecting the child. Shortly after asking the help of the league, the father ran away, taking with him the child whom the mother left alone in the house while she went to the "movies." The women who assert themselves in their new rights are in a small minority. A young Polish woman complains that the women of her group are too submissive even in this country, and "bear beatings just as their mothers did in the old country." In the great majority of foreign-born families, as in all families, the question of the legal rights of the woman is never raised. The habits and attitudes formed under the old system of law and customs are carried over into the life in the new country, and are changed so gradually and imperceptibly that no apparent friction is caused in the family group. Moreover, in many cases where the woman perceives her changed position she is able to make her husband see it too, and she herself is able to work her way through to a new understanding. It is interesting to note that the women of the foreign-born groups who have worked their way through are now bending their energies toward helping the women who have not yet started.
- 65. III THE CARE OF THE HOUSE The work that the housewife must do in the care of the house is the maintenance of such standards of cleanliness and order as are to prevail. It includes the daily routine tasks of bedmaking, cooking, sweeping, dusting, dishwashing, disposing of waste, and the heavier work of washing, ironing, and periodic cleanings. NEW HOUSEKEEPING CONDITIONS The foreign-born housewife finds this work particularly difficult for many reasons. In the first place, housekeeping in the country from which she came was done under such different conditions that it here becomes almost a new problem in which her experience in the old country may prove of little use. The extent to which this is true varies from group to group. To understand the problems of any particular group, careful study should be made of the living conditions and housekeeping practices in the country from which it came. Some of the women with whom we have conferred have described housekeeping as they knew it in the old country. These descriptions are suggestive of the character of the change and the difficulties involved. Mrs. P., a Polish woman from Posen, for example, said that: Houses in the village in which she lived were made of clay, with thatched roofs, clay floors, and about ten feet high. They were made in rows, for four families or two families, with one outer door opening from a hall into which the doors from all the dwellings opened. Each dwelling had one small window, and a fireplace. Water was out of doors. In the four-family house there
- 66. were two chimneys. The outside door did not open into the road. FLOOR PLAN OF HOUSES IN POLAND The floors were covered with sand, and new sand was put on when the room was cleaned. The fireplace had a hook from which hung the kettle, and in one corner was the oven, a little place set off by a board covered with clay. Walls were whitewashed. Mrs. P. said that the housework is much more difficult in this country, with the cleaning of woodwork, washing windows, care of curtains, carpets, and dishes, and more elaborate cooking. In the old country the family washing was done only once a month, except in cases where there were small children. Then it was done weekly; and if the family lacked sufficient clothing, the washing had to be done oftener. There the meal was one dish, from which the entire family ate; here there is a variety of food and each person has his own plate and eating utensils, so that even the dishwashing is a greater task. In coming to this country many women do not see that the windows need washing or that the woodwork should be cleaned, etc.
- 67. The beds were made of boards covered with straw, not as a straw mattress. Sheets were laid over the straw to make it softer. Each person had two pillows, very large and full, so that they sleep in a "half sitting" position. Feather beds are used for warmth, and no quilts or blankets were known in the old country. Lithuanian women, likewise, have pointed out that at home most of the women worked in the fields, and that what housekeeping was done was of the simplest kind. The peasant house consisted of two rooms, one of which was used only on state occasions, a visit from the priest, a wedding, christening, or a funeral. In summer no one sleeps in the house, but all sleep out of doors in the hay; in winter, women with small children sleep inside, but the others sleep in the granary. Feather beds are, in these circumstances, a real necessity. Thus the bed that is found in this country is unknown in Lithuania, and the women naturally do not know how to care for one. They not only do not realize the need of airing it, turning the mattress, and changing the bedding, but do not even know how to make it up properly. Other processes of housekeeping—dishwashing, scrubbing, and washing—prove equally difficult, and it is said that most of the women do things in the hardest possible way, chiefly because the processes are different here and they lack the technique to do their work in the easier way. Naturally, too, when work in the fields has occupied most of their time, they lack also habits of order and routine in their household tasks. The Italian women, especially those from southern Italy and Sicily, have also spoken of their difficulties in housekeeping under new conditions. In Italy the houses, even of the relatively well-to-do peasants, were two-room affairs with earthen floors and little furniture. The women had little time to give to the care of the house, and its comfort and order were not considered important.
- 68. The experience in doing the family washing is said to typify the change. In Italy washing is done once a month, or at most, once a fortnight, in the poorer families. Clothes are placed in a great vat or tub of cold water, covered with a cloth on which is sprinkled wood ashes, and allowed to stand overnight. In the morning they are taken to a stream or fountain, and washed in running water. They are dried on trees and bushes in the bright, Italian sunlight. Such methods of laundry work do not teach the women anything about washing in this country, and they are said to make difficult work of it in many cases. They learn that clothes are boiled here, but they do not know which clothes to boil and which to wash without boiling; and as a result they often boil all sorts of clothing, colored and white, together. In Italy washing is a social function; here it is a task for each individual woman. DEMANDS OF AMERICAN COOKERY Cooking in this country varies in difficulty in the different national groups. In the case of the Lithuanians and Poles, for example, the old-country cooking is simple and easily done. Among others it is a fine art, requiring much time and skill. The Italian cooking, of course, is well known, as is also the Hungarian. Among the Bohemians and Croatians, too, the housewives are proverbially good cooks and spend long hours over the preparation of food. Croatian women in this country are said to regard American cookery with scorn. They say that Croatian women do not expect to get a meal in less than two or three hours, while here all the emphasis is on foods that can be prepared in twenty or thirty minutes. It is not always easy to transplant this art of cookery, even if the women had time to practice it here as they did at home. The materials can usually be obtained, although often at a considerable expense, but the equipment with which they cook and the stoves on which they cook are entirely different. The Italian women, for example, cannot bake their bread in the ovens of the stoves that they use here. Tomato paste, for example, is used in great quantities by Italian families, and is made at home by drying the tomatoes in
- 69. the open air. When an attempt is made to do this in almost any large city the tomatoes get not only the sunshine, but the soot and dirt of the city. The more particular Italians here will not make tomato paste outdoors, but large numbers of Italian families continue to make it, as can be seen by a walk through any Italian district in late August or early September. In general, in the groups in which cooking was highly developed, a great deal of time was devoted to the preparation of food. If the housewife wishes to reduce her work in this country, she finds that some of the ingredients which make our cooking simpler are unknown to her. The Bohemians, for example, do not know how to use baking powder, and the same is true of the women in Lithuanian, Polish, and Russian groups, where the art of cooking is less developed. With this lack of experience in housekeeping under comparable conditions, the foreign-born housewife finds the transition to housekeeping in this country difficult at best. As a matter of fact, however, the circumstances under which she must make the change are often of the worst. She is expected to maintain standards of cleanliness and sanitary housekeeping that have developed with modern systems of plumbing and facilities for the disposal of waste that are not always to be found in the districts in which she lives. Even a skillful housewife finds housekeeping difficult in such houses as are usually occupied by recently arrived immigrants. WATER SUPPLY ESSENTIAL In the first place, there is the question of water supply. Cleanliness of house, clothing, and even of person is extremely difficult in a modern industrial community, without an adequate supply of hot and cold water within the dwelling. We are, however, very far from realizing this condition. In some cities[13] the law requires that there shall be a sink with running water in every dwelling, but in other cities even this minimum is not required. The United States Immigration Commission, for example, found that 1,413 households
- 70. out of 8,651 foreign-born households studied in seven large cities, shared their water supply with other families. Conditions have improved in this respect during the last decade, but it is a great handicap to efficient housekeeping if water has to be carried any distance. Further inconvenience results if running hot water is not available, which is too often the case in the homes of the foreign born. THIS PUMP SUPPLIES WATER TO FOUR FAMILIES Cleanliness is also dependent, in part, upon the facilities for the disposition of human waste, the convenient and accessible toilet connected with a sewer system. These facilities are lacking in many immigrant neighborhoods, as has been repeatedly shown in various housing investigations. For example, in a Slovak district in the Twentieth Ward, Chicago, 80 per cent of the families were using toilets located in the cellar, yard, or under the sidewalk, and in many cases sharing such toilets with other families. One yard toilet was used by five families, consisting of twenty-eight persons.[14] The danger to health, and the lack of privacy, that such toilet accommodations mean have been often emphasized. In addition, it
- 71. enormously increases the work of the housewife and makes cleanliness difficult, if not impossible. There is also the question of heating and lighting the house. Whenever light is provided by the oil lamp, it must be filled and cleaned; and when heat is provided by the coal stove, it means that the housewife must keep the fires going and dispose of the inevitable dirt and ashes. In the old country the provision of fuel was part of the woman's duties; and in this country, as coal is so expensive, many women feel they must continue this function. Here this means picking up fuel wherever it can be found—in dump heaps and along the railroad tracks. A leading Bohemian politician said that he often thought, as he saw women prominent in Bohemian society, "Well, times have changed since you used to pick up coal along the railroad tracks." OVERCROWDING HAMPERS THE HOUSEWIFE The influence of overcrowding on the work of the housewife must also be considered in connection with housekeeping in immigrant households. That overcrowding exists has been pointed out again and again. Ordinances have been framed to try to prevent it, but it has persisted. In the studies of Chicago housing a large percentage of the bedrooms have always been found illegally occupied. The per cent of the rooms so occupied varied from 30 in one Italian district to 72 in the Slavic district around the steel mills. The United States Immigration Commission found, for example, that 5,305, or 35.1 per cent, of the families studied in industrial centers used all rooms but one for sleeping, and another 771 families used even the kitchen. Crowding means denial of opportunity for skillful and artistic performance of tasks. "A place for everything and everything in its place," suggests appropriate assignment of articles of use to their proper niches, corners, and shelves. One room for everything except sleeping—cooking, washing, caring for the children, catching a breath for the moment—means no repose, no calm, no opportunity for planning that order which is the law of the well-governed home.
- 72. Yet there is abundant evidence that many families have had to live in just such conditions. The housework for the foreign-born housewife is often complicated by other factors. One is the practice to which reference has been already made of taking lodgers to supplement the father's wages. In discussing this subject from the point of view of the lodger, it has been pointed out that the practice with reference to the taking of boarders and lodgers varies in different places and among different groups. The amounts paid were not noted there, but they become important when considered together with the service asked of the housewife. Usually the boarder or lodger pays a fixed monthly sum— from $2 to $3.50, or, more rarely, $4 a month—for lodging, cleaning, washing, and cooking; his food is secured separately, the account being entered in a grocery book and settled at regular intervals. Sometimes the lodger does his own buying, but the more common custom is to have the housewife do it. Occasionally he does his own cooking, in which case payment for lodging secures him the right to use the stove. More rarely, as in some of the Mexican families visited in Chicago in 1919, he is a regular boarder, paying a weekly sum for room and board. Just what keeping lodgers means in adding to the duties of the housewife can be seen from the following description of the work of the Serbo-Croatian women in Johnstown, Pennsylvania:[15] The wife, without extra charge, makes up the beds, does the washing and ironing, and buys and prepares the food for all the lodgers. Usually she gets everything on credit, and the lodgers pay their respective shares biweekly. These conditions exist to some extent among other foreigners, but are not so prevalent among other nationalities in Johnstown as among the Serbo- Croatians. In a workingman's family, it is sometimes said, the woman's working day is two hours longer than the man's. But if this
- 73. statement is correct in general, the augmentation stated is insufficient in these abnormal homes, where the women are required to have many meals and dinner buckets ready at irregular hours to accommodate men working on different shifts. The Serbo-Croatian women who, more than any of the others, do all this work, are big, handsome, and graceful, proud and reckless of their strength. During the progress of the investigation, in the winter months, they were frequently seen walking about the yards and courts, in bare feet, on the snow and ice-covered ground, hanging up clothes or carrying water into the house from a yard hydrant. WOMEN WORK OUTSIDE THE HOME Another factor that renders housekeeping difficult is the necessity of doing wage-paid work outside the home, to which reference has already been made. Women interviewed have repeatedly emphasized the difficulties that this practice creates in connection with the housekeeping. A recent study of children of working mothers, soon to be published by the United States Children's Bureau, carried on at the Chicago School of Civics and Philanthropy, obtained the testimony of the mothers as to the difficulties involved. This study showed that in many cases the household duties could not be performed at the proper time; 60 women, for example, of the 109 reporting on this question, said that they did not make their beds until night; 105 said their dishes were not washed after each meal, but in 41 cases were washed in the mornings, and in 56 not until night. Three washed them in the morning if they had time, and five left them for the children, after school. Many women who worked outside the home did their housekeeping without assistance from other members of the family. This meant that they had to get up early in the morning and frequently work
- 74. late at night at laundry or cleaning; 49 women, for example, washed in the evening; 25 washed either Saturday, Sunday, or evenings. HOUSING IMPROVEMENT Enough has probably been said to show that the work of caring for the house under the conditions existing in most immigrant neighborhoods, is unnecessarily difficult for the foreign-born housewife. The most obvious point at which these burdens might be lightened so that the housewife could have time for other duties, would appear to be through improvement of housing. With an awakened realization of this fact, both on the part of the foreign- born woman herself and the community of which she is an inevitable part, will come the solution of these difficulties. A protest, however inarticulate or indirectly expressed by her, will find its response in a growing realization that plans for improvement must be developed. The several housing projects that have already been offered are suggestive of the problems and possibilities along this line rather than useful as hard-and-fast solutions. They not only meet the needs of the more inadequate immigrant housing conditions, but provide improvement upon most native-born conditions. In this connection interest naturally centers on the war-time housing projects of the United States government, on the experiment of the Massachusetts Homestead Commission at Lowell, and on certain enterprises carried out by so-called limited dividend companies. The first two are especially interesting, in that they recognize that supplying houses to the workers is not a function that can be wholly left to private initiative. It is not possible to discuss these projects in detail, nor is it necessary.[16] It is sufficient to consider them here with reference to the contributions they might make in helping the immigrant housewife. In the first place, they provide for a toilet and a bath in every house, and a supply of running water that is both adequate and convenient. In the matter of kitchen equipment there is an attempt to provide some of the conveniences. Both provide a sink
- 75. and set wash-tubs equipped with covers. They must be set at a minimum of thirty-six inches from the floor in the United States plans. Both make provision for gas to be used for cooking, although the coal stove is accepted. The kitchens in the Massachusetts houses are also provided with kitchen cabinets, with shelves under the sink, and with a drain for the refrigerator. In other ways also consideration for the housewife is evidenced. Electricity is urged for lighting, passages through which furniture would not go are avoided, the size of the living room is adapted to the sizes of the most commonly purchased rugs, etc. Study of the Massachusetts plans reveals other interesting features, such as the care given to the location of the bathroom and the attention to the size of the doors, so that the mother at work in her kitchen can watch the children at play in other rooms. Both projects are interesting also in that they realize the necessity of a "front room" or parlor, and prescribe a minimum number of bedrooms—three in the Massachusetts, and two in the United States experiment. Both require closets in every bedroom wide enough to receive the men's garments on hangers, and rooms of such size that the bed can stand free of the wall and out of a draught. It is evident that the plans for houses in both projects provide very definite improvements in the matter of the conveniences to which the immigrant is not accustomed in the houses at present available to him. Some limitations, however, become apparent by comparing them with the recommendations of the Women's Subcommittee of the Ministry of Reconstruction Advisory Council, England. That committee emphasizes the importance of electricity for lighting, and urges "that a cheap supply of electricity for domestic purposes should be made available with the least possible delay." The American plans agree that electricity is the preferred lighting, but gas is accepted by the United States government, although not by the Massachusetts plan. There is no suggestion of developing a cheaper supply of electricity.
- 76. The English women also suggest the desirability of a central heating plant as a measure that would lessen the work of the household, afford economies in fuel, and render a hot-water supply readily available. They urge, therefore, further experimentation with central heating. The American plans have no suggestions to make at this point, but accept the coal stove or the separate furnace in the higher-priced houses as the means of heating. While they provide for hot water, no suggestions are made as to how this is to be supplied. It is presumably done by a tank attached to the range, which means that hot water is not available when there is no fire in the range; that is, in summer and during the night. It should also be noted that these plans make no suggestions for co-operative use of any of the equipment of the household. There is another point at which the architects and builders failed to take sufficient notice of the problem of lightening the women's work —namely, in their attitude toward the separate family home as compared with the multiple family dwelling. The Massachusetts Commission was, by the terms of the Act creating it, limited to the provision of one or two-family houses; the United States government standards were definitely against the building occupied in whole or in part by three or more families. Tenement and apartment houses are considered generally undesirable, and will be accepted only in cities where, because of high land values, it is clearly demonstrated that single and two-family houses cannot be economically provided, or where there is insistent demand for this type of multiple housing. This judgment, however, has by no means met with universal approval. Those architects who think in terms of the woman's time and strength consider the merits of the group and of the multiple house. For example, those who planned the Black Rock Apartment House Group in Bridgeport, Connecticut, the open-stairway dwellings, the John Jay dwellings on East Seventy-seventh Street, New York City, and the Erwin, Tennessee, development, maintain
- 77. that the advantages of the separate house in privacy, independence, and access to land can be secured by the multiple arrangement. Not only can economies in the use of the land be practiced, but protection and assistance for the women and children can be obtained, and there is the possibility of devices for convenient and collective performance of many tasks. It is unnecessary to review the arguments for the one or for the other. It is evident that the group house, and perhaps the multiple house, offer such inducements in the economy of space and the possibility of assigning areas of land to definite and anticipated uses, that their further adaptation to family needs must be contemplated. It is generally assumed that the family group wants the separate house. The question of interest for this study is one of the desire of the immigrant groups in this respect. Their preference should be an indispensable element in the formulation of housing standards. There is not, however, a great deal of evidence on this subject. The fact that immigrants live in the city in the congested districts may only indicate that they have had no choice in the matter. Most of the officers of certain immigrant building and loan associations interviewed for this study thought there was a preference for the single-family dwelling when it could be afforded. That also is the belief of the investigators in this study, who think that the use of multiple houses indicates not the immigrants' desires, but their acceptance of what is before them, and that the dream of almost every immigrant family is to have a house of its own, to which is attached a little garden. How far the desire for the separate house is confused with the desire for the garden would be difficult to say. It is certain, however, that in general the immigrant has known only one way to have the garden, and that was by having a separate house. There is universal agreement that especially the foreign-born family desires access to land for whose cultivation they may be responsible, and whose produce both in food and in flowers they may enjoy. Recently, however, certain architects have been interested in working out
- 78. Welcome to our website – the perfect destination for book lovers and knowledge seekers. We believe that every book holds a new world, offering opportunities for learning, discovery, and personal growth. That’s why we are dedicated to bringing you a diverse collection of books, ranging from classic literature and specialized publications to self-development guides and children's books. More than just a book-buying platform, we strive to be a bridge connecting you with timeless cultural and intellectual values. With an elegant, user-friendly interface and a smart search system, you can quickly find the books that best suit your interests. Additionally, our special promotions and home delivery services help you save time and fully enjoy the joy of reading. Join us on a journey of knowledge exploration, passion nurturing, and personal growth every day! ebookbell.com