Reading comprehension in the English language is a process that has been studied from different disciplines, due to the need for certification of the English language as a second language; this process is required in the qualification processes of students enrolled in a postgraduate course. Therefore, in this paper, a new model is proposed, based on the versatility of first order logic, situation calculus and semantic relationships to model the passages of this type of certification. The model is represented by a knowledge base, which considers the cognitive model of Van Dijk and Kinstch.
Reading requires the development of a complex cognitive system that supports the
processing of information at different levels, whether conscious or unconscious. A good
reader is one who is able to construct an integrated mental representation of the text,
which is also coherent and accurate [
Readers of texts in languages other than their native one have two challenges: first
to translate to their native language, and then to map the structure from the vocabulary
that they know of the foreign language [
In the TOEFL (Test of English as a Foreign Language), in particular the reading
comprehension section, in order to answer the questions, the reader builds a model of
knowledge representation, which requires applying inferential processes to understand
the meaning of the text [
For this reason, it is important to pose models of knowledge representation of the passages of the reading comprehension section, with their corresponding questions and answers. The purpose is to establish the meaning of the text according to the context in which it is found.
Although first order logic allows us to model assertions or predicates, it is also important to establish a model of the contexts in the passages, so situation calculus is a useful tool to do so. Another important aspect to favor this representation of knowledge is the identification of the semantic relationships present in the passages. So, depending on the type of relationship, inferences can be produced that help establish strategies to respond appropriately to the questions of the reading comprehension section of the TOEFL.
The content of this paper is divided as follows: section 2 shows a theoretical framework of cognitive models of reading comprehension and inferential processes. Section 3 shows the semantic relationships, the calculation of situations and their relation to the modeling of predicates considering the context. In section 4, an example is presented of how a passage is converted to a knowledge base according to semantic relationships and situation calculus. Finally the conclusions and future work are presented in section 5. 2
Cognitive models are characterized by studying how human beings know, think and remember. It explores the capacity of human minds to modify and control the way in which stimuli affect behavior and sustains learning as a process where meanings are modified internally.
This is achieved by integrating the mechanisms of short and long term memory
with those of inference and although they are performed automatically, not all human
beings perform it at the same time or in the same way. As Johnson[
Reading comprehension is the process of simultaneously extracting and
constructing meaning with the following objectives: to decipher how letters represent words,
to accurately and efficienly translate letters to sounds (extract meaning from text), to
formulate a representation of the information that is being presented, which inevitably
requires the elaboration of new meanings; and to integrate the new information with
the old (construction of meaning) [
The precursors of these mental models were Van Dijk and Kintsch[
Two key concepts in this recall process were the ’macrostructure’ and the
’superstructure’, which were proposed in that investigation. This theory assumed that textual
processing is done in cycles, due to the limited capacity of short-term memory after
decoding the code, and that a representation of the text (base text) in the memory was
gradually built up in this way. This base text not only consists of a connected sequence
of ’propositions’, but also establishes a hierarchical structure of ’macro propositions’,
which correspond to the most important and least important themes of the text deduced
(inferred) by the reader[
The base text, then, results from sequences of propositions that are made coherent
by the ’repetition of arguments’. The macro structures, on the other hand, can be
defined as higher order propositions that include underlying propositions. In other words,
macropropositions are constructed with the micropropositions of a text, and are a
summary or other abstract structure underlying a text, so they must be inferred from the
text. Thus the micro and macropropositions form a ’macrostructure’ of the text, that is,
a semantic structure that defines the overall meaning of a text. However, these
structures must associate with a context associated with the reader’s experience. This forms
situational model, which is a cognitive model of the situation reflected in the text that
contains inferred material[
Another proposed model is that of Miller[
Also in 1995, the 3CAPS model was proposed by Goldman, Varma and Cote[
Later Kintsch[
Considering the recovery of memory to support the inference process, in 2001,
Singer[
On the other hand, Lin and Patel [
Even though several cognitive models have been proposed, the Kinstch and Van Dijk model has interrelated elements that fuse cognitive psychology and predicate logic for support in the process of reading comprehension, which is interesting to address from the point of view of nonmonotonic reasoning. 2.1
Garc´ıa Garc´ıa[
In the case of the decoding process, this consists of deciphering a code; in this case it is about giving a meaning to the printed letters. Two decoding processes are allowed: associating the written word with the available meaning in the subject’s memory and recoding what it means to transform the printed letters into syllables and into sounds in order to activate the meaning.
Literal comprehension consists of combining the meaning of several words in an appropriate way to form propositions and adheres to the information explicitly reflected in the text.
Inferential comprehension provides a deeper understanding of the text and goes beyond what is explained in it. The reader, through inferences, elaborates a more integrated and schematic mental representation from the information expressed in the text and from his previous knowledge.
Metacomprehension is the consciousness and control that the reader has of his process of understanding. It consists of establishing some goals for reading, checking if they are being reached and rectifying quickly if necessary.
As can be seen, the process of reading comprehension is complex and although there are theoretical models and computational approaches, it is necessary to identify the elements that hinder the construction of meaning that in this case are the inferences, since they are applied according to the level of experience of the reader and the strategies used to build new meaning from the existing one.
An inference is considered a conclusion or opinion that is reached taking into account
evidence or known facts. From the point of view of logic, an inference is defined as the
process of deriving logical consequences from assumptions [
They fulfill diverse functions in the speech processing, allow the identification of
referents, disambiguate and / or complete semantic meanings, as well as emptiness
substitute when the information is not explicitly available. Linguistic inference has been
defined as: ”any conclusion that a reader reasonably derives from a sentence”
(Hurford and Heasley 1983)[
Within linguistic inferences, Kempson[
There are several types of inference, as mentioned in the works of Leo´n and Pe´rez
[
Semantics is the part of linguistics that studies the meaning of words, sentences and
expressions of the language. All the words that maintain a relationship of meaning
between them are part of the same semantic field. For example: carnation and rose
belong to the semantic field of flowers [
Among the words that form a semantic field can exist relations of hyponymy and
hyperonymy, synonymy and antonymy. [
A word is a hyponym of another if its meaning is included in it. For example, a rose
is a hyponym of flower. A word is a hyperonym of another if its meaning includes the
meaning of it. For example, flower is a hyperonym of rose.
Criterial
Degree of probability
vs.Certain [
Synonymy is a semantic phenomenon by which the same concept or idea can be
expressed with two or more different words. The synonymous words have, therefore,
an equal or very similar meaning within the same context[
Antonymy is a semantic phenomenon that occurs when two words have an opposite meaning, e.g., bad and good.
These semantic relationships are present in the texts and in most cases their
identification supports the inferential strategies to answer the questions of reading
comprehension. However, to map the meaning of the texts to a knowledge base requires modeling
these relationships from two aspects that depend on the context[
When a context is not required, but have a vocabulary of terms is available that
allows the reader to determine hyponyms and hyperonyms, in this case, the entailment can
be used to express these relationships, as shown in Zenteno[
According to Zenteno, entailment is also identified as ’inference’, as proposed by
Kempson (1977) [
Because of its ambivalent nature, entailment can be defined, logically, in terms of
valid rules of inference or, semantically, in terms of the assignment of truth or falsehood
of related propositions: ”(a proposition) p semantically entails (a proposition) q if and
only if in every situation where p is true q is also true (or in all the worlds where p is
true, q is true) ”(Levinson 1983: 174) [
If entailment is handled as a logical relationship between propositions expressed by
sentences, this idea has made it possible to relate systematically (with reference to
predicate calculus and predicate relations, such as symmetry, transitivity, and reflexivity)
notions such as hyponymy, synonymy, antonymy, related opposition and contradiction.
Thus, for example, Palmer (1981) points out that hyponyms, predicates that establish
a relationship of meaning, such that the meaning of one is included in that of another,
involve entailment: for example, rose implies flower. The lexemes that are associated
through a hyponymy relationship can also establish transitivity: rose implies flower, and
flower implies being alive [
There are other types of entailment between propositions, for example, in the following propositions: David killed Golitat, Goliath died. The relationship is valid considering killing(David, Goliath) can be sure that if this proposition is true implies that dying(Goliath) is also true even if there are not hyponymy relations, but rather a cause with an effect.
While semantic relationships can be modeled with entailment, the context that is fundamentally required in synonymy has not yet been considered, so modeling the context will be addressed in the next subsection. 3.3
To address the calculation of situations, first the preliminary elements of first-order logic are defined below:
A first order language with equality is specified by two disjoint sets of symbols
called the vocabulary of the language (Reiter)[
– Equality =.
Parameters these vary with interpretation – Quantifier symbol : 8. – Predicate symbols: For each n
or n-ary predicate symbols. – Function symbols: for each n or n-ary function symbols.
0, a set (posibly empty) of symbols, called n-place 0, a set (posibly empty) of symbols, called n-place
Terms, atomic formulas, literals, well formed formulas are defined as usual, as are the concepts of free and bound ocurrences of a variable in a formula. A sentence is a formula with no free variables. The symbols _; ^; 9 are defined to be suitable abbreviations ocurrences of a variable in a formula.
Assume given a nonempty set I, whose members are called sorts, in this case those terms are defined: Logical symbols: As before, except that for each sort i, there are infinitely many variables x1i; x2i; : : : of sort i. Each term is aasigned a unique sort, as follows:
2. If t1; : : : ; tn are terms of sort i1; : : : ; in respectively and f is a function symbol of sort < i1; : : : ; in >, then f (t1; : : : ; tn) is a term of sort in+1
1. When t and t are terms of the same sort, t = t is an atomic formula 2. When P is an n-ary predicate symbol of sort < i1; : : : ; in > and t1; : : : ; tn are terms of sort i1; : : : ; in respectively, then P(t1; : : : ; tn) is an atomic formula.
The calculation of situations is a logical formalism designed for the representation and reasoning about dynamic domains. It was proposed by John McCarthy in 1963.
Baker mentions that it allows to represent changing scenarios as a set of formulas of
first order logic. Reiter[
It is also defined the function do(a; s) that denotes a successor situation for s to execute the action a. For example, the predicate cause(virus; disease) indicates the action that a virus causes a disease on the object x to the object y. If the function do (put (A, B), s) is applied, it means the situation resulting from putting A in B when the situation s occurs. In the calculation of situations, actions are denoted as function symbols and situations are first order terms.
With this scheme, the logic of first order can be used to formalize the effects of various actions.
– two function symbols of sort situation:
A constant symbol S0, denoting the intial situation
A binary function symbol do : action ? situation ! situation – A binary predicate symbol @: situation ? situation, defining an ordering relation on situations – For each n 0, countably infinitely many predicate symbols with arity n, and sorts (action [ ob ject)n. These are used to denote situation independent relations like virus(Rotavirus);
Reiter[
The Projection Problem Suppose D is a basic action theory, a1; :::; an is a sequence of ground action terms, and G(s) is a formula with one free variable s, whose only situation term is s. Determine whether: D j= G(do([a1; :::; an]; S0)).
For example, a projection query for the sequence of actions Closest meaning(proven, showed) might be: The word ”proven” is closest meaning of showed over virus theories D j= is(do(Closest meaning(proven; showed); virus theories))).
To return results of this projection, Reiter[
In case of the evaluation of querys in the database, the response to a query W answers to the projection query in the resulting situation.
These definitions of the calculation of situations, it is possible to extend to formalize
the context. McCarthy[
assert(query(K; f); f yes) – Frame Axiom (propositional): if f is an closed formula, and yes does not ocurr in the context Y then assert(K; Y) assert(query(K; f); Y) – Interpretation Axiom (Qualitative) if x is free unique variable in f then
assert(query(K; f(X )); f(X ) answer(X )) – Frame Axiom (Qualitative) if x is free unique variables in f and answer does not ocurr in the context Y, then assert(K; Y) assert(query(K; f(X )); Y) – Answer Axiom : assert(reply(K; f); Y) assert(K; f Y)
If this approach is possible to build applications that improve the training in the section of reading comprehension and thus the rates of upgrading of use of the TOEFL.Considering this support theory, a knowledge base is modeled below by means of a TOEFL type passage. 4
– First, TOEFL Reading tests what specific facts are mentioned in the passage, as well as what is not mentioned. The typical format is the good old ”One Best Answer”. An effective strategy is to make a ”road map” of the passage right away so that you can find the answers more efficiently. Certain skills such as skimming and scanning will help you more efficiently establish this map. – Second, they test about certain pronouns, like ”its” or ”their”, refer to in specific parts of the text. – Finally, they’ll ask what inferences can be made form certain information.
In the reading passages [
The following example of passage is available at the University of Chuvanan[
In the process of reading comprehension of these passages there are implicit inferences to be made according to the situational model of each reader, as Van Dijk commented in 1978. It is therefore necessary to identify semantic relatios as: synonyms, hyponyms, hyponyms, antonyms, cause-effect relations and entailment. In Table [?] the knowledge base about passage is represented.
The content of the passage allows to establish predicates and hyponims and hyperonims can be identified, in the case of these structures it is not required to establish a context to define them, but the entailment is present in this form: – hyponim(human, livingBeings) means human entail livinBeings. – hyponim(virus, parasites) means virus entail parasites.
In the case of synonyms, in this case, they are not present in the content of the passage.
Once the passage is represented, it is necessary to process the questions and identify the context to answer them considering the answer options as shown in Table 3: Before microbes were discovered It was believed that some diseases were caused by assert(sinonym(emanate, (A) germ-carrying insects (B) certain strains of bacteria release),context) (C) foul odors released from swamps where context=swamps (D) slimy creatures living near swamps
So in this case, it is required to answer the question, according to the context of disease, these querys can be modeled by calculating situations, for example, the modeled assertion: assert(synonim(proven; showed); disease) imply that a answer will be found, in this case it is necessary to prove that proven is a synonim of showed, applying the axiom of qualitative interpretation, this result is assert(query(K; synonim(proven; showed); disease); disease answer(X )) where f(X ) would represent the predicate of synonymy and the answer X is showed, discarding the others.
In the same case, in the assertion assert(sinonym(emanate; release); swamps), the terms emanate and release are not similar unless they are related in the context of swamps. 5
Although reading comprehension is a complex process, designing representation models that allow the identification of terms, semantic relationships, entailments and contextrelated assertions will favor the generation of inferences to design query-answer systems to improve the achievement in the reading comprehension sections of TOEFL exams.
The assertions generated from the calculation of situations provide intuitive expressiveness to associate the semantic relations to a context, so this representation will favor to identify properties and enrich inferential processes.
Kinstch and Van Dijk’s model, emphasize the situational model as an element dependent on the reader’s experience, with the description of the contexts generated from the calculation of situations, it is possible to generate a representation closer to the reader’s experience,thus strategies can be elaborated to improve the process of reading comprehension.
This work was carried is supported by the Sectoral Research Fund for Education with the CONACyT project 257357 and Thematic Network in Language Technologies(RedTTL) with its program Research stays for students.