Explanation Based Learning: Fundamentals and Applications

By Fouad Sabry

What Is Explanation Based Learning

A type of machine learning known as explanation-based learning, or EBL for short, takes advantage of an extremely robust, or even flawless, domain theory in order to generalize from training data or construct concepts. It also has a connection with encoding, or memory, which assists with learning.

How You Will Benefit

(I) Insights, and validations about the following topics:

Chapter 1: Explanation-based learning

Chapter 2: Computational linguistics

Chapter 3: Natural language processing

Chapter 4: Corpus linguistics

Chapter 5: Parsing

Chapter 6: Question answering

Chapter 7: Link grammar

Chapter 8: Grammar induction

Chapter 9: Structured prediction

Chapter 10: Deep linguistic processing

(II) Answering the public top questions about explanation based learning.

(III) Real world examples for the usage of explanation based learning in many fields.

(IV) 17 appendices to explain, briefly, 266 emerging technologies in each industry to have 360-degree full understanding of explanation based learning' technologies.

Who This Book Is For

Professionals, undergraduate and graduate students, enthusiasts, hobbyists, and those who want to go beyond basic knowledge or information for any kind of explanation based learning.

• Language: English
• Publisher: One Billion Knowledgeable
• Release date: Jun 30, 2023

Book preview

Chapter 1: Explanation-based learning

In order to generalize or form concepts from training examples, explanation-based learning (EBL) is a type of machine learning that makes use of a very strong, or even perfect, domain theory (i.e. a formal theory of an application domain analogous to a domain model in ontology engineering, not to be confused with Scott's domain theory).

A chess-playing computer program is a great illustration of EBL with a perfect domain theory. Many irrelevant features, such as the unique dispersion of pawns on the board, are part of a certain chess position that has a significant feature like Forced loss of black queen in two moves. EBL is able to learn from a single training example by extracting the most important elements.

If a domain theory contains all that might possibly be needed to answer any question in that domain, then it is perfect or complete. The chess rules alone constitute the domain theory for this game. If you know the rules, you should be able to figure out what to do in any given circumstance. Combinatorial explosion, however, makes it impossible to actually make such an inference. EBL employs training examples to make it easier to efficiently search for the logical consequences of a domain theory.

Each training example in an EBL system is ultimately deduced from the system's pre-existing database of domain theory. Having a concise proof of the training case expands the domain-theory database, making it much easier for the EBL system to discover and classify similar future examples. Minton examined the method's primary drawback: the increasing difficulty and time required to apply the learnt proof macros.

There are four inputs to the EBL program:

the domain of hypotheses (the set of all possible conclusions)

a theory of domains (axioms about a domain of interest)

examples for training (specific facts that rule out some possible hypothesis)

criterion for operation (criteria for determining which features in the domain are efficiently recognizable, e.g. which features are directly detectable using sensors)

Natural language processing is an area where an EBL shines (NLP). Here, a treebank is used to fine-tune a rich domain theory, i.e. a natural language grammar, for a specific application or language usage (training examples). In an early step, Rayner eliminated the basic language (domain theory) and used specific LR-parsing techniques to solve the utility problem, achieving massive speedups at the expense of coverage but with an improvement in disambiguation. Surface generation, the inverse of parsing, has also been explored using EBL-like approaches. Both a goal coverage/disambiguation trade-off (= recall/precision trade-off = f-score) and the entropy of the treebank's or-nodes can be used to infer or. In addition, from generic unification grammars, EBL can be utilized to create grammar-based language models for speech recognition. Take note of how the utility problem first identified by Minton was resolved by abandoning the original grammar/domain theory, and of how the papers cited often use the term grammar specialization (the polar opposite of the original term explanation-based generalization) in their own writings. Data-driven search space reduction could be a good name for this method. Guenther Neumann, Aravind Joshi, Srinivas Bangalore, and Khalil Sima'an are only a few of the persons who have contributed to the field of EBL for NLP.

{End Chapter 1}

Chapter 2: Computational linguistics

An interdisciplinary discipline, computational linguistics focuses on the computer modeling of natural language, as well as the investigation of relevant computational methods to various linguistic challenges. In general, computational linguistics draws on a wide variety of fields, including but not limited to linguistics, computer science, artificial intelligence, mathematics, logic, philosophy, cognitive science, cognitive psychology, psycholinguistics, ethnography, and neuroscience.

In the past, computational linguistics developed as a subfield of artificial intelligence carried out by computer scientists who had specialized in the use of computers in the translation and analysis of natural languages. During the 1970s and 1980s, the subject was able to become more established thanks to the introduction of independent conference series as well as the foundation of the Association for Computational Linguistics (ACL).

The Association for Computational Linguistics (ACL) provides the following definition for the field of computational linguistics::

...the use of scientific methods and computer analysis to the study of language. Researchers in the field of computational linguistics are interested in developing computer models of many different types of language processes.

The terms natural language processing (NLP) and (human) language technology are increasingly being seen as being almost synonymous with the word computational linguistics. This is the case in the year 2020. Since the beginning of the 2000s, these phrases have placed more of a focus on the investigation of real applications rather than theoretical concepts. Although they solely pertain to the subfield of applied computational linguistics, in practice, they have largely supplanted the term computational linguistics in the NLP/ACL community. This is because they refer more explicitly to the subject of applied computational linguistics.

The study of computational linguistics incorporates both theoretical and practical aspects. The field of theoretical computational linguistics focuses on problems that arise in the fields of cognitive science and theoretical linguistics.

The creation of formal theories of grammar (parsing) and semantics is an important part of theoretical computational linguistics. These theories are often rooted in formal logics and symbolic (knowledge-based) techniques. Research domains that are within the purview of theoretical computational linguistics include the following::

The computational difficulty of natural language, which is based mostly on automata theory and makes use of context-sensitive grammar and linearly bounded Turing machines.

Determining appropriate logics for the encoding of linguistic meaning, automatically creating such logics, and reasoning with those logics are all components of computational semantics.

Machine learning, which has typically relied on statistical approaches and, from the middle of the 2010s, neural networks: Socher et al., is the most important aspect of applied computational linguistics (2012)

Other divisions of computational into main fields according to various criteria exist, such as the divide that exists between theoretical and practical computational linguistics. These divisions of computational include::

regardless of the spoken or written form of the language that is being processed: The fields of voice recognition and speech synthesis