搜索
您的当前位置:首页正文

心理逻辑理论:心理语言假设和推理角色语义学的实证研究范式

来源:二三娱乐
2013年11月 外国语 November 2013 第36卷第6期 Journal of Foreign Languages Vo1.36 No.6 文章编号:1004—5139(2013)06—0027—08 中图分类号:HO30 文献标识码:A 心理逻辑理论:, 理语言假设和推理角色语义学的实证研究范式 David P.O’Brien ,李淑静 (1.纽约市立大学柏鲁克分校,纽约10010;2.河南大学外语学院,河南开封475001) 摘要:据Fodor[15]和Fodor&Pylyshyn[12]提出的“心理语言假设”认为,思维是以心理语言进行的;心理 语言有一个由逻辑句法和逻辑语义组成的表征格式。据此而论,思维可以看作是这些表征的句法运算。本文 认为,这一系统的语义表征可以很好地由推理角色语义学来说明,而Braine和O’Brien[3]的心理逻辑理论 可以为心理语言运行的实证研究提供一个研究范式。 关键词:心理语言假设;推理角色语义学;心理逻辑理论 Mental Logic Theory:A Paradigmatic Case for Empirical Research on the Language of Thought and Inferential Role Semantics David P.0’Brien ,LI Shujing (1.Baruch College and the Graduate Center of the City University of New York,New York,NY 10010,USA 2.College of Foreign Languages,Henan University,Kaifeng 475001,China) Abstract:The Language of Thought Hypothesis,which follows from proposals of Fodor[15]and Fodor and Pylyshyn [12],proposes that thinking takes place in a mental language,which has a representational format consisting both of a logical syntax and a logical semantics.Thinking thus can be conceived of as a set of syntactic operations on such representations.We argue that the semantics of such a system is best provided by an inference role semantics and that the mental logic theory of Braine and O’Brien[3]can provide a paradigmatic case study in how the empirical inves— tigation of the language of thought can proceed. Key words:Language of Thought Hypothesis;inferential role semantics;mental logic theory We begin with the fact that human cognition includes a declarative memory,that is,memory for information in propositional form.This is the sort of information that students provide when responding to exam questions llke:Wbat is the capital city of Henan Province?Who was the first President of the Democratic Republic of the Congo?How many kilometers are there between Shanghai and Hong Kong? What is the largest prime number under 2,000 7 It is also the kind of information involved when a man facing divorce asks for directions to his lawyer’s office,or when someone discusses the implications of taking a job in another city,or asks a friend for the best way to prepare a dish that one hasn’t cooked before.The responses to all of these questions are provided in linguistic form.and they involve decisions about which answers,or parts thereof,are true,or might be false,which allows people to dispute one another over such things.Two interlocutors might argue,for example,about whether the capital of Henan Province is Kaifeng or Xi’an,for example,or whether 1,999 is a prime number or not. 27 We take it as axiomatic that in order to record something propositionally,that is,to record it in a declarative memory,there must be a format with which to record it——without a format for recording something,nothing can be recorded.A representational format adequate to recording information about the world requires some logical structure.It would need some sort of predicate-argument structure to distinguish between properties and the entities that have those properties and it would need to distinguish between relations among objects and among the objects that are related———it would need some sort of sentential operators.It would need to record alternatives among objects and among properties held by those objects。。——the sort of thing done by a disjunction operator _——and it would need to record when multiple objects hold the same property or multiple properties are held by an object。・—_—the sort of things done in logic by a conjunction operator.It would need to be able to record whether a property is absent rather than present,or when a propositional claim is false——things that are done in logic by negation.Further,if a person’s thinking is to go beyond the information given,as Bruner[7]expressed it,one needs some way of representing that a proposition is being treated as thoulgh it were tue fror the sake of constructing a line of reasoning under it,even though one is not claiming that it is true——the sort of thing done in logic by supposition in constructing conditional propositions.As Maenamara[30]suggested,such a system would also need referring devises such as indexicals,e.g,using English words such as this and that——the linguistic equivalent to using one’s index finger to point out to which object or objects one is referring, and quantifiers,which can be conveyed with words like all,some,many, ,etc.,but also with various idiomatic expressions like as many as lfeas on a dog’s back. The arumentg we are making here is similar to proposals by Fodor[1 5]and Macnamara[30],and it suggests a need for something innate in the representational format,so one should expect that some essential parts of the representational structure are innate and thus universal___——and it suggests that those inferential forms that preserve truth in lines of reasoning that are applied easily in one language ought to be available for easy application in other languages.We adopt the term“language of thought’’for the representational format(e.g.Fodor[13;14;16];Fodor&Pylyshyn[12];Davies[9];Lycan[29]; Jacob[21];Markic[31];Harman[18];Field[10];see also Rey[39;40;41]),although in speaking of a language of thought,we are not committing to Fodor’s claim that the content predicates of the language of thought are innate(see also Jackendoff[20]). We thus follow Braine and O’Brien[3]in distinguishing between a“language of thought”and a “syntax of thought.’’In Fodor’s conception,both the content predicates and the syntax of the language of thought would be innate,but in our conception the syntax would be innate although the content predicates would be acquired.It is the innate syntax that provides the basic representational system and it would have to include a mental logic to which we were referring above as the format for representing propositional knowledge that makes declarative memory possible. From our perspective,the content predicates are acquired,and together the innate part and the acquired part constitute a language of thought in which thinking and reasoning take place.The concept of a mental logic thus is intimately connected to the concept of a language of thought.To represent propositions and their inference mechanisms,some system of representation is needed,and to move beyond the mere recording of observation or information in discourse,the language of thought needs some inference mechanisms to produce new propositions from those already recorded in declarative memory. 28 Indeed,because many propositions cannot be simple representations of what is perceived directly from the world,that is,they go beyond representations of simple qualia,many propositions could not be held in declarative memory at all unless they were the output of inference・making processes.Consider the existence of a conditional proposition such as if a natural number ends in zero,it is divisible by five. Could the truth of this proposition be determined by empirical presentation or sheer declaration?To know that it is true is to know that it follows from what we know about the domain of the natural numbers.One could consider the smallest such number 10,and realize that it is the addition of ive taken twifce,and the subsequent such number 20,is also obtained by the addition of 10,so it,too,is divisible by five,as is the next subsequent such number 30,and so forth.Since it follows that all subsequent natural numbers ending in zero can be obtained in the same way,it follows that all natural numbers ending in zero have this property of being divisible by five.The proof of this particular proposition follows from something logicians call mathematical induction,and without something inferential,one never could assert that this conditional proposition is tue.So,the language of trhought needs more than simple passive procedures for storing propositions that record qualia,or are verbatim copies of propositions received in discourse processing.If the language of thought is to include propositions that go beyond the atomic,it must include inference procedures that produce propositional theorems. So,the fact of declarative memory propels us to propose that human cognition includes(a)a language of thought that includes some predicate—argument structure,and this stuctrure enables representation of entities and the properties that attach to the entities,(b)some sentential operators, which enable representation of such things as alternatives,conjunctions,negations,and conditionals,(c) some indexical and quantifying terms,which enable representation of which entities and which properties are being related to one another,and(d)some inference—making procedures that enable one to go beyond the information given and to produce propositions beyond the atomic.Further,features(a)to(d)of this language of thought constitute in large part the innate syntax of the language of thought,although,unlike Fodor,we allow that the content predicates in the language of thought reasonably are obtained through experience rather than being innate;we do not propose that a predicate like“is creating a theory of quantum computing”can reasonably be claimed as an innate predicate. The language of thought hypothesis carries a model concerning where thinking takes place.Consider a participant in a reasoning experiment,who is presented with some premise assumptions for example,in English or in Mandarin,and is required to provide some propositional response.The implication of the language of thought hypothesis is that these premise propositions are translated from the natural language in which they are presented to the participant(e.g.English or Mandarin)into the language of thought,and it is in the language of thought that the sentences obtain propositional meaning,and where inferences that follow from these propositions -——taken together with other propositions available from long—term memory‘_—— are made using the inference procedures of the language of thought.For the participant to make the required response,the inferences obtained in the language of thought are then translated into the natural language in which the discourse is being conducted. This notion of obtaining an understanding of the meaning of received sentences through translation into the language of thought(and of providing output by translation from the language of thought to a natural language)can be useful in understanding what is involved in translation between two natural 29 languages,for example,when someone is translating from English to Mandarin or vice versa.Clearly, translations between natural languages cannot be accomplished by mere word—for—word translation———— looking up each English word to find its corresponding Mandarin word.Note that the basic units of two languages often are not the same,and although one reasonably could claim that word is the basic category for English,it is difficult to find an exactly corresponding category for Mandarin——typically one word in English can require several characters in Chinese,and the character seems reasonable as the basic unit of meaning in Chinese,so the basic units of the two languages do not correspond comfortably.Moreover,the surface—structure grammars of the two languages do not correspond easily——Chinese,for example,is far less inflected.The language of thought hypothesis allows a translator not to be daunted by these inconveniences,however,as the translator takes the English sentence and(automatically)translates it into the language of thought,and then translates the language of thought——and the understanding of meaning that occurs only there———to the surface structure of Mandarin.When someone is acquiring a new language,for example,living in a different country or taking a foreign language course,one often is asked whether one is yet able to“think in the new language.’’or whether one“dreams in the new language yet,”and these are taken as equivalent indications of how thoroughly one is acquiring the new language. Although these two questions can appear to be essentially the same,they are quite different when viewed from the language of thought hypothesis.In our view of the hypothesis,one never is thinking in any natural language—either one’s mother tongue or a second language that is acquired later in life.One thinks in the language of thought regardless of whether discourse is taking place in one particular natural language or another.What changes over time as one gets increasingly proficient in a new language is that the effort required to translate between the new language and the language of thought diminishes’__——the process of translation between the natural language and the language of thought becomes increasingly automatic.Thus,although dreaming in a new language may increase as one becomes more proficient with the new language,thinking never will be taking place in a natural language—acquired or one’s mother tongue. either one that is newly 1.Some Considerations About Logical Semantics and the Lexical Entries for the Vocabulary of a Mental Logic To get an idea of why we think about the logical semantics for a mental logic in the way that we do, consider the following two examples of the sorts of logical inferences that we think are of a basic sort in the mental logic.In the first example,you are told that all of the children in the neighborhood went fishing today.You know that Mang Yin is one of the neighborhood children,so you infer that Mang Yin went ishifng today.Suppose instead that you are told that some of the neighborhood children went fishing.You know that Yun Fat is one of the neighborhood children.In this case,however,you withhold the inference that Yun Fat went fishing.Indeed,understanding this difference in when such an inference can be made is tantamount to understanding the meanings of the two quantifiers all and some,and someone who does not appreciate which of these inferences should be made and which should be withheld could not be said to understand the difference in meanings between the two quantifiers.For this reason we propose an inference—role approach to logical semantics,that is,to provide a semantics for logic particles such as and,or,not,and/f,and for logic quantiifers such as all and some.The inference—role approach to 30 semantics proposes that meaning of a term is provided by the pattern of inferences that it sanctions,and philosophical advocates of this approach to semantics generally(e.g.Gentzen[17];Quine[38];Block [1];Peacocke[37];Loar[28];Woods[43];Jackendoff[20];Fodor and Lepore[11];Field [1 0];Brandom[6];see also Wittgenstein[42])have indicated that logical connectives are the paradigmatic cases of how meaning can come from patterns of implication.We are not necessarily advocating this approach for the lexicon of the language of thought as a whole(see Fodor and Lepore [1 1],for a discussion of limitations to an inference—role semantics),but we see no reason that it cannot provide a semantics for logical connectives,and it’s difficult to see how a different approach to the semantics of logic terms can be as successfu1. An older notion of meaning,growing principally from the philosophical writings of British empiricism, is that meaning concerns a correspondence between words and the world.As Fodor and Lepore[1 1]put it,the word dog means what it does rather than proton because it is associated with a mental image that means dog rather than proton,and the mental image means dog because it resembles dogs a lot,but protons not at al1.But on this analysis,the Morning Star and the Evening Star should have the same meanings because they both refer to the same object,even though they may not mean the same thing at al1.The alternative is to seek semantic meaning not in correspondence between word and world,but in the inferences that are sanctioned by the word. For a word like shirt or red there is some appeal to the notion that the meaning of the word can be found in some sort of correspondence between word and something in the world(or at least in a mind’s eye view of the world).One can have an image of redness,or some sort of collection of related images,of experiences of colors that relate to the word red,or that correspond to the word shin——or at least to the concepts that correspond to the lexical items’__——and one can imagine a compositional proposition of the two words in a conjunctive proposition to provide meaning to the phrase red shin.But what are the correspondences between word and world,or between the concept linked to the lexical item and the world ofr a propositional logic operator such as or or To what would an expression like/fp then q correspond in the world? Standard logic provides truth tables to deal with the semantics of sentential operators like/f,or,and, etc.,with the truth table assignments defining the objects to which the meanings of the words are presumed to correspond.The tuth table meanings of logirc particles,however,are notoriously problematic.The truth table for conditionals has been traced to Philo of Megara in approximately 300 B. C.E(Kneale&Kneale[25]).This way of viewing conditionals,often referred to as the material conditional,or as material implication,can be represented as follows: P true q true P or q true false true true false true false false false true which makes/fp then q tuer unless P is tue and q irs false.Thus,on this way of viewing the semantics of conditionals,/fp then q is tuer whenever P is false.Imagine an innocent college student who had just ifnished an introductory logic course being stopped by the police and asked,if you were on Main Street last night,then you murdered the young woman,didn’t you?Remembering the truth table for conditionals,he answers yes,because he knows the antecedent to be false,thus making the conditional true,and thus implicating himself in the murder.What he says to the police can and will be used against him.Not surprisingly,much logic literature through the years has concerned itself with how to extract logic from such paradoxical effects of the material tutrh table. A single truth table simply is lacking when one attempts to use a truth table to provide the semantics of or,that is,the operator associated with the concept of alternatives.Two distinct truth table assignments have been associated with or,one that treats or as inclusive: p true q true tue rP or q true true true false true false false false false and one that treats or as exclusive: p tue rg tue rP or q false true false true false true false false true false Ironically,it appears thus that one needs to introduce some sort of logic particle for alternatives into the semantics of alternatives if one is to use truth tables to provide a semantics for or,that is,one needs to define the semantics of or as allowing either the first or the second of these tuth trables. Given the obvious problems for the truth tables to provide a semantics for the ordinary way in which logic particles are used,one might wonder why they play such an important role in symbolic logic and are almost ubiquitous in logic textbooks.The answer may be straightforward:they are useful in constructing metalogical proo ̄of soundness and completeness for formal logic systems.Among those working in philosophical linguistics who take the language of thought hypothesis seriously,however,there is unanimity in not suggesting truth tables as an account of the semantics of logic particles.Instead,the laternative to what Fodor and Lepore[1 1]called the“Old Testament”semantics of correspondence has been an inferential—role semantics,which Fodor and Lepore referred to as the“New Testament’’of linguistic semantics.An inference role semantics views the meaning of a word as being founded in the inferences that it supports,which often are discussed in terms of what a cognitive scientist might call a semantic network.So the word shirt,for example,might lead to such propositional inferences as“is a garment,…‘worn on the upper body,’’“typically has two sleeves,”and so forth,as well as negative inferences such as“is not a type of building,’…is not a pair of pants,’’“is not a hat,”and so forth. For logical terms such as/f,not,and,or,all,some,etc.,an inferential-role semantics as part of a psychological theory of mental logic needs to specify empirically testable inference procedures,which the theory of Braine and O’Brien does,especially given the theory’s coordination of schemas with a reasoning routine that guides the application of schemas to construct lines of reasoning.(We discuss the schemas and the reasoning routine in more detail later.1 32 Consider how the mental logic theory proposed by Braine and O’Brien[4;3])proposes that judgments about the truth of conditional propositions,that is,of the form if P then q,are made.The proposal includes a schema for conditional proof and an inference procedure that applies the schema when a conditional is to be evaluated(e.g.as tentative conclusion to a reasoning problem).The schema for conditional proof holds that to assert a proposition of the form if P then q,first suppose P.For any proposition q,that follows from the supposition of P taken together with other background premise assumptions,one can assert ifP then q. ̄SWhen a tentative conclusion of the form ifP then q is presented to be evaluated as true or false,the reasoning routine adds P to the pl’emises and then reasons towards q as the conclusion to be evaluated.When q is derived,then fiP then q is judged true.When the negation of q is derived,then ifP then q is judged false.Of particular interest is that this latter prediction of a“false” judgment differs from what would be predicted if people were using the turth table for the material conditional to provide a logical semantics for conditionals(the“false”judgment would be withheld because P ml’ght be false).Our proposal thus is that the meaning of logic terms like/f,and,not,or,all, some and so forth,are captured by the procedures that determine which inferences are sanctioned using these terms 2.The Theory of Mental Logic and Some Empirical Evidence That Supports it Braine and O’Brien[3]proposed three empirical routes which could contribute information about the nature of the mental logic and the syntax of thought.The first is through reasoning,and it focuses most directly on the basic inference forms and representational system of the mental logic.The second is through universals of language and the third is through what is semantically and syntactically most primitive in language acquisition;both of the latter two focus on the syntax of thought,i.e.on the nature of the representati0nal system,more directly than on inference. We address here only the first route,which is the one on which we have worked principally,and we present it as a paradigmatic case study of how the empirical investigation of the syntax of thought can proceed. Braine and O’Brien[3]presented a detailed hypothesis about the mental logic both at the level of a sentential logic——that is,for sentential operators like if,and,not,and or——and at the level of a predicate calculus——that is,including further description of the internal structure of propositions that include predicate—argument structure and its related quantifiers.To illustrate the inference schemas at the sentential level,we list some here,with an example for each that refers to content about toy animals and rfuits in a box: (a)pl OR...OR pn;~pi. .pl OR...OR pi一1 OR i+1 OR...OR pn There is a tiger or an elephant in the box;There is not a tiger in the box ‘..There is an elephant in the box (b)pl OR...OR pn;IF pl THEN q;...;IF pn THEN q There is a rabbit or a hen in the box: ①Braine and O’Brien also described some constraints on which propositions may be used in arguments constructed under a supposition.We do not have the space to discuss them here,except to note that any proposition entered into a line of reasoning under a supposition would need to still be ture if the supposition is ture. 33 If there is a rabbit in the box,then there is a banana in the box; If there is a hen in the box.then there is a banana in the box . .There is a banana in the box  .ql OR...OR qn (C) pl OR...OR pn;IF pl THEN ql;...;IF pn THEN qn.There is an elephant or a monkey in the box; If there is an elephant in the box then there is a pineapple: If there is a monkey in the box then there is a banana ’..There is a pineapple or a banana in the box (d) IF P THEN q;P.’.q If there is a cat in the box then there is a monkey in the box: There is a cat in the box 。..There is a monkey in the box The schemas at the predicate—logic level are representationaUy more complex because, as we wrote above,they require further analysis of the internal composition of propositions.Our description of the representational format of the predicate logic differs in some important ways from what one finds in a standard logic textbook.To illustrate,consider the following propositional representation in standard logic: (V x)(Px or Qx), which would correspond in English to For all . is P or is Q.This expression illustrates that the quantifiers in standard logic have outside scope and are content genera1.What we mean by this is that(a) the quantifier is placed outside the main body of the proposition rather than being attached directly to the elements within the expression,and(b)the content of the predicates that are quantiifed are not attached to the quantifiers,but are found only in a backwards reference to the quantifier when encountered.In the mental logic representation,one finds inside scope and the quantifiers are content speciic:f S [All X]OR S [PRO—All X] which rendered in English would says that All of the X’s satisfy Sl or they satisfy S2.Note that this way of representing the universally quantiied proposition has insifde scope and the quantiier is contentf speciic.This ifs accomplished by having the argument terms—the X’s~related by the use of a pronoun in the second clause,that has an anaphoric reference to the specified nouns in the first clause.This use of pronouns to keep track of the relationship between the argument terms and the predicates to which they are attached is much more natural,we believe,and closer to the sort of arrangement one finds typically in natural languages,which tend to have inside scope and content speciifcity(see Ioup[1 9]),and so the format seems to us to be much more natural that than of standard logic. We illustrate the sorts of inference schemas that Braine and O’Brien[3]proposed at the predicate logic level with the following partial list(for details of the notational system and a complete list of schemas,see Braine&0’Brien[3],Chapter 11): (e)S,[All X];S:[All X]/S [All X]0&S [PRO—All X] The boys wore blue jcans;The girls played with the boys/The boys wore blue jcans and the girls played with them. (f) S [All X]OR S [PRO—All X]/S [All x:NEG S [PRO]) The boys either played with girls or fought with girls/The boys who did not play with girls fought with girls. (g)NEG E[~Some X:S PRO]&S:[PRO]~ ;S [a];[a] The boys that played with Mary did not wear sandals. [X]/NEG S [a] There were no boys who wore sandals and blue jcans;The boys that played with Mary wore blue jcans/ (h)S [All X]OR S [PRO—All X];S3[All X:S1[PRO]I;S [All X:S2[PRO] /S,[A11 X] AI1 the ears in the lot have stickers or the guards tow them away;The cars that have stickers are Toyotas;The cars that the guards tow away are Toyotas/All the cars in the lot are Toyotas. The reasoning program that applies the schemas to construct lines of reasoning has been presented both for the sentential and predicate—logic levels.It distinguishes between two levels of implementation. First,there is a direct—reasoning routine(DRR)that applies the schemas when their requisite conditions have been met.For example,when both/f p then q and P are held at the same time in working nlemory, Schema(d)above will be applied,leading to the inference q.Likewise,when P or q and not,p are held conjointly in working memory,Schema(a)above will be applied,and q will be inferred.(The DRR for the predicate—logic level also includes provisions for topics to guide the application of schenlas;we have insuficifent space available here to describe this feature and refer the reader to Braine&0’Brien『3], Chapter l1 for details.) Braine and O’Brien[3]argued that the larger set of schemas,together with the DRR,provide a basic syntax of the language of thought and thus will be applied with minimal effort(so long as a person is attentive),will appear early in development,should appear across languages and cultures and should not require literacy or education.They also described some reasoning strategies that go beyond what is available on the DRR.One such strategy,called the supposition of alternatives strategy,can be applied when a premise set contains a disjunction,and each disjunct in turn can be treated as a supposition in order to apply the schema of conditional proof to derive a conditional with the disjunct as its antecedent, feeding into a line of reasoning using(c)or(d)above.Another strategy,called the enumeration of alternatives a priori,can be applied when a premise set contains a conditional of the orfm/fp then q;the reasoner can add the proposition P or not P to the premise set,again feeding into lines of reasoning using schemas that apply to alternatives such as(c)and(d).A third strategy is the reductio ad absurdum,in which one supposes a proposition in the service of finding a contradiction in a line of reasoning under the supposition in order to falsify the supposition.There is evidence that although these strategies are not always available,they are common among university students and can be found in other populations. 3.Some Empirical Evidence Supportive of the Mental Logic Theory Mental logic theory has predicted successfully which reasoning problems people solve and which they do not,the perceived relative difficulties of the problems solved,and the orders in which intermediate inferences are made as a problem is thought through.It also has shown that the basic inferences are made routinely and effortlessly in text comprehension,that those inferences are made online during text comprehension as the information enters working memory,and it has shown that the orders in which inferences are made depend on when the schemas become available and not on the order in which premise information is presented. We turn first to some problems on which the theory makes predictions that differ from those that would 35 be made if people were reasoning using standard logic of the sort in a typical logic textbook.As nmntioned above,people do not respond as they should if they were using truth tables when they are making judgments about when eortditionals are false.When/f P then q is being evaluated,the melqtta logic procedure adds P to the set of premises and then reasons towards q.When q is derived, P then q is judged tue,butr when not q is derived the conditional is judged false—a judgment that does not olflow in standard logic because P might be false.Braine,Reiser,and Rumain[2]reposed that aduhs make the sorts of judgments predicted by mental logic theory,and this finding has been extended to children by O’ Brien,Dias,Roazzi,and Braine[33].Consider the four following problems: (i)There is either a cat or a hat in the box;therefore if there is not a hat there is a cat (j)There is either a cat or a hat in the box;therefore if there is not a hat there is not a cat (k)There is not both a cat and a hat in the box;therefore ifthere is a cat there is not a hat (1)There is not both a cat and a hat in the box;therefore if there is a cat there is a hat Problems(i)and(k)should lead to“tue”rresponses,both on standard logic and on mental logic, and both children and adults give these responses.In Problems(J)and(1)only mental logic theory predicts false responses,and this is the response that both children and adults make.Recently,O’Brien, Roazzi,and Athias[35]found the same pattern of responses on similar problems presented to an illiterate and uneducated indigenous population in a remote area of Brazilian Amazonas.②Clearly.participants of all ages and populations evaluated conditionals in the way predicted by the procedures described in mental logic theory.③ The most basic prediction of the theory。。——that those inferences that follow front application of the DRR will be made routinely,whereas those inferences requiring strategies that go beyond the DRR will be made far less often,has been supported by results in Braine et a1.[32]and O’Brien et a1.[35]. Braine et a1.presented to university students problems that were solvable by the DRR and compared them to problems that required additional reasoning strategies.O’Brien et a1.made the same comparison in remote villages in the Brazilian Amazon with both literate and educated adults and illiterate and uneducated adults.The problems all presented content that would not invite responses based on wodd knowledge(e.g.1etters on an imaginary blackboard,bracelets and baskets of various colors).Almost no errors were made on the direct—reasoning problems,but,as was expected,problems that required more sophisticated reasoning strategies led to far more errors for all populations.Solution rates thus could be predicted by knowing whether a problem required reasoning strategies or just the DRR. Braine et a1.[2],with sentential problems,and Braine et a1.[5],with predicate level,provided an additional sort of evidence to support the claim that not only were the direct reasoning problems being solved,but that they were being solved in the way described by the theory.Participants were directed to rate the perceived relative dificulfty of each problem on a Likert—type scale,and regression models were ②In 0’Brien et a1.(1998)and O’Brien et a1.(2011)a kind of training was given to prompt“can’t tel1”responses, which would be the predicted response on standard logic.The concern was that children and non—educated participants might be resistant to providing such responses.Eyen after training,such responses were not given on these problems. ③There are other inferences that would follow if one were reasoning using standard logic that would not be accepted in mental logic.For example,P or q could be inferred from P,but people seem not to judge such inferences as sound,as predicted by mental logic theory. 36 constructed from the perceived—dificulty ratifng data that assigned a weight to each schema.This enabled prediction of the difficulty of each problem(as equal to the sum of the weights of each schema required for problem solution in the lines of reasoning as predicted by the DRR).On sentential—level problems, correlations between predicted and observed difficulties accounted for 66%of the variance(53%with problem length partialed out),even when the weights were obtained with one set of problems and the observed ratings were obtained with another set of problems and different participants.On predicate—level problems,the ratings predicted by the schema weights again were correlated highly with the observed rating(69%of the variance;56%when problem length was partialed out),even when observed ratings came from new problems and different participants than those used to generate the schema weights. Lea et a1.(1990)[26]and Lea[27]provided evidence for the use of the sentential schemas in text processing,as did 0’Brien,Roazzi,Dias,and Soskova[34]for the predicate—logic schemas.Both sets of studies reported that the core inferences are made routinely when their premises are embedded within short story vignettes and that these inferences are made so easily that people usually do not realize that any inferences are being made at al1.Further,the mental logic inferences are unlike other sorts of inferences made while reading,such as inferences from story grammars,scripts,etc.,in that the mental logic inference are made whether or not they are needed in order to maintain textual coherence.Included among the text-・comprehension studies were online measures of reaction times investigating sentential・・level schemas(Lea[27])and predicate—level schemas(0’Brien,et a1.[34]).In one story version presented by Lea[27]the reader was told that Mary will wear her black dress if a party is a Halloween party,and Mary discovers that it is a Halloween party;in a control version Mary does not discover whether or not it is a Halloween party.When subjects subsequently were presented a naming task on which they were to say the word“black”when it was presented o13 a computer screen,they responded faster following the former story than the latter,showing that inferences can be measured online. O’Brien et a1.[32]and Braine et al[5]provided problems that required multiple inferences in a line of reasoning,and participants made the intermediate inferences in the orders predicted by the theory. Participants were asked to write down what they figured out in the order that they figured them out.The orders in which the premises were presented were varied,testing the prediction that the order in which participants would write down their reasoning steps would correspond to the order in which the schemas became available and would not vary as a function of the order in which the premises were presented. As an example,consider the following problem pair from O’Brien et a1.[32]that referred to letters written on an imaginary blackboard: Prob1em 1 N or P Not N If P then H Prob】em 2 Not both Z and S If H then Z If P then Z 工f H then Z Not both Z and S. Not N N or P On Problem 1 the first two premises allow derivation of P.This then combines with(c)to derive H, which,when premise(d)is read,allows derivation of Z,which allows derivation of not S when premise (e)is read.Thus,as each new premise is read,it fuliflls the conditions to trigger a schema.Problem 2 37 presented the same premises in the reverse order,so the DRR is unable to apply any of the core schemas until all of the premises have been read.Participants wrote down inferences in the same order Ofl the two problems,as predicted,showing that infeFences are nlade as the schetnas become available. 4.Concluding Remarks We do not clainl that the empirieal evidence is dispositive that mental logic theory exclusively and exhaustively captures the inferenee—making procedures of the language of thought.It does provide, however,a detailed set of descriptions of。how propositional information is represented and what inferences I’eadily can be drawn from those representations,and its predictions stand up extremely weII to empirical assessment.The only other approach that has attempted the same sort of scope across logic particles and quantifiers has been menta1 models theory(e.g.Johnson—Laird&By rne[22;23]and we shal1 discuss below reasons 1o think that mental logic theory provides the more plausible candidate for the basic syntax of the language of thought. Over and Evans【36]have proposed an account of conditionals that in some ways is similar to ours. Like ours,it proposes that the first step in evaluating a conditional is to suppose its antecedent and then to reason towards its consequent,using probability procedures to make evaluations of conditionals.As yet, however,thei r。proposal is limited to conditionals and says nothing about the broader set of logic operators such as alternatives and has not provided a description of a representational format that goes beyond conditionals,which is a principal requirement for a psychological theory to work if it is to provide a descriptiOil of the language of thought. The most comprehensive ahernative theory is the mental—models theory of Johnson—I—aird and hi s colleagues(e.g.Johnson—I aim&Byrne[22;23]),because it addresses conditionals,negations, alternatives,and conjunctions,and has some proposals about reasoning with quantiifers.There are some aspects to the theory,however,that weaken it as a competing candidate.For one thing,it specifically excludes the use of variables,using only tokens to represent individual objects. The claim that propositions are represented and inferences are made without anything logical, including the use of any variables,is problematic.Johnson—Laird and Byrne[24]wrote that although the models do not contain variables,variables do occur“1’11 the initial semantic representations”from which models are constructed[24:376].They wrote that quantifiers provide“the raw material for a reeursive loop that is used in building or manipulating a mode1.Thus,the universal quantifier‘all’elicits a recursion that deals with a set exhaustively,whereas the existential quantiier‘some’elifcits a recursion that does not”(Johnson—Laird&Byrne[23:178]),and they provided the following example:all x’s are equal to the sum of some y and some z,which first is parsed to yield: (m) (All x)(Some Y)(Some z)(x=Y+z) A model is constructed by giving some arbitrary value to the first variable term in the equation(x), some arbitrary value to the second term(Y),and,because the degrees of freedom have been exhausted, a constrained value is assigned to the third term in the equation(z).This procedure loops over the equation several times,recording the output on each pass,resulting in a model like the following: (n) [8 6](1 6 4 2)(7 7 2 2 4 4) in which both of the numbers in the first set are equal to sortie number in the second set plus some number 38 ln the third set. As Johnson—Laird and Byrne wrote,“the procedure is analogous to the standard interpretation of quantifiers in the predicate calculus except that it constructs a model of the assertion.”[23:1 80]They are correct that the representation in(m)and the procedure that generates(n)~with its arbitrary instantiations—include quantifiers and variables.The model in(n)that results from the procedure, however,does not convey the meaning of the expression from which it was generated,and reasoning uses only the mode1.Note that the expression in(m)could not be retrieved from the model in(n).The model merely provides a couple of numbers,each of which is the sum of two other numbers from the adjacent sets,but in no way does the model in(n)capture the meaning of universality of the expression in(m), nor the relations among the variables x,Y,and z in(n1).The square brackets around[8 6]do not indicate universality;they merely indicate that“once the set is complete no further items can be added to it”[23:1 80],and thus the square brackets do not convey the meaning of All x’s bearing a particular arithmetic relation to some Y’s and z’s.Nor is there any indication in the model that infinitely many other models could be constructed.A model for such an expression is merely a set of instantiations that are consistent with that expression,and does not itself have any logical structure with which one can know which information can be integrated,and how,or which,inferences can be drawn.Put simply,the semantic meaning from which the model is constuctred is not contained in the model that finally is constructed.If reasoning really takes place at the level of such models.then it must be taking place devoid of crucial information that had been contained at the input level of the pre—models representations. We provide this example only to illustrate why we think that our theory provides a candidate that is psychologically more plausible as an account for the logical syntax of the language of thought.A representational format without variables hardly is adequate to keep track of properties and the entities that hold those properties,which,as we wrote in the beginning of the article,is what the syntax of language must provide.We invite the reader to investigate the two theories in detail to decide the relative merits of the two proposals keeping this in mind. We end by returning to requirements of the language of thought as described by Fodor and Pylyshyn [1 2],who argued that the representational system must have a syntax and semantics in which molecular representations are built up from atomic constituents,and the content of the combinations is a function of the semantic content of the atomic constituents together with the syntactic structure.Further,the inferential operations on these representations are causally sensitive to the syntactic structure as defined by the combinatorial syntax.We believe that the representational format and the inferenee procedures of the mental logic we have described fit these demands exactly,and our theory thus provides an excellent case study in how the empirical study of the language of thought hypothesis can proceed. References: [1]Block,N.Functional role and truth conditions[J].Proceedings of the Aristotelian Society LXI,1987:157— 181. [2]Braine,M.D.S.,Reiser,B.J.&B.Rumain.Some empirical justiifcation for a theory of natural proposi— tional logic[A].In Bower,G.H.(Ed.),The Psychology ofLearning and Motivation:Advances in research and thinking[C].New York:Academic Press.1984,(18):317—371. 39 [3]Braine,M.D.S,&D.P.O’Brien(eds.).Mental Logic[C].Mahwah,NJ:Lawrence Erlbaum Associates, 1998. [4]Braine,M.D.S.&O’Brien,D.P.A theory of if:A lexical entry,reasoning program,and pragmatic princi— ples[J].Psychological Review,1991,(98):182—203. [5]Braine,M.D.S.,O’Brien,D.P.,Noveck,I.A.,Samuels,M.,Lea,R.B.,Fisch,S.M.,&Y.Yang. Predicting intermediate and muhiple conclusions in propositional logic inference problems:Further evidence for a mental logic[J].Journal ofExperimental ychology:General,1995,(1 24):263—292. [6]Brandom,R.Articulating Reasons.An Introduction to Inferentialism[M].Cambridge,MA:Harvard Universi— ty Press,2000. [7]Bruner,J.Beyond the Information Given[M].New York:W.W.Norton and Company,1973. [8]Davies,M.Concepts,connectionism,and the language of thought[A].in Ramsey,W.,Stich,S.P.&D. E.Rumelhat(edsr.).Philosophy and Connectionist Theory[c].Hillsdale,NJ:Lawrence Erlbaum,1991. 229—257. [9]Davies,M.Connectionism,modularity,and tacit knowledge[J].British Journal for the Philosophy of Sci— ence,1989,(40):541—555. [10]Field,H_Logic,meaning and conceptual role[J].Journal fPhiolosophy,1977,(69):379—408. [11]Fodor,J.A.&E.Lepore.Why meaning(probably)isnt conceptual role[J ?Mind and Language,1991, 6(4):328~343. [12]Fodor,J.A.&z.W.Pylyshyn.Connectionism and cognitive architecture:a critical analysis[A].In Pin— ker,S.&J.Mehler(eds.).Connections and Symbols[c].Cambridge,MA:MIT Press,1988.3—72. [13]Fodor,J.A.LOT 2:The anguageL f)厂Thought Revisited[M].Oxford:Oxford University Press,2008. [14]Fodor,J.A.Psychosemantics:The Problem,ofMeaning in the Philosophy ofMind[M].Cambridge,MA: MIT Press,1987. [15]Fodor,J.A. e Language ofThought[M].New York:Thomas Y.Crowell,1975. [16]Fodor,J.A.The Modularity ofMind[M].Cambridge,MA:MIT Press,1983. [17]Gentzen,G.Investigations into logical deduction[J].American Philosophical Quarterly,1964,(1):288— 306.(Original work published in 1935). [18]Harman,G.Thought[M].Princeton,NJ:Princeton University Press,1973. [19]Ioup,G.Some universalsfor quantifiertype[A].InKimball,J.(ed.).Syntax&Semantics(Vo1.4)[c]. New York:Academic Press,1975.37—58. [20]Jackendoff,R.Semantic Structures[M].Cambridge,MA:MIT Press,1990. [21]Jacob,P.What Minds Can Do:Intentionality in a Non—Intentional World[M].Cambridge,UK:Cambridge University Press,1997. [22]Johnson—Laird,P.N.&R.M.J.Byrne.Conditionals:a theory ofmeaning,pragmatics,and inference[J]. Psychological Review,2002,(109):646—678. [23]Johnson—Laird,P.N.&R.M.J.Byrne.Deduction[M].Hillsdale,NJ:Psychology Press,1991. [24]Johnson—Laird,P.N.&R.M.J.Byrne.Mental models or formal rules?[J].Behavioral and Brain Sciences, 1993,(16):368—380. [25]Kneale,W,&M.Kneale.The Development ofLogic[M].Oxford,England:Clarendon Press,1962. [26]Lea,R.B.,OBrien,D.P.,Fisch,M.,Noveck,I.A。,&M.D。S.Braine.Predicting propositional logic inferences in text comprehension[J].Journal ofMemory&Language,1990,(29):361—387. [27]Lea,R.B.Online evidence for elaborative inferences in text[J].Journal ofExperimental Psychology:Learn— ing,Memory,and Cognition,1995,(21):1469—1482. 40 [28] Loar,B.Mind and Meaning[M].Cambridge,MA:Cambridge University Press,1981. Lyean,W.G.A deductive argument for the representational theory of thinking[J].Mind and Language, [29] 1993,8(3):404—422. [30] Macnamara,J。A Border Dispute:The Place oflogic inpsychology[M].Cambridge,MA:MIT Press,1986. [31] Markic,O.Is language of thought a conceptual necessity?[J].Acta Analytica,2001,16(26):53—60. O’Brien,D.P.,Braine,M.D.S.&Y.Yang.Propositional reasoning by models?Simple to refute in princi— [32] ple and in practice[J].Psychological Review,1994,(101):71 1—724. 0’Brien,D.P.,Dias,M.G.,Roazzi,A.&M.D.S.Braine.Conditional reasoning:The logic of supposi— [33] tion and children’S understanding of pretense[A].In Braine,M.D.S.&D.P.O’Brien(Eds.).Mental Logic[C].Mahwah,NJ:Lawrence Erlbaum Associates,1998.245—272. O’Brien,D.P.,Roazzi,A.,Dias,M.G.,&J.Soskova.Prevendo infer ̄neias 16gico—predicativas no pro— [34] cessamento de texto[J].Revista Interamericana de Psicologia(Interamerican Journal of Psychology),2007, 41(2):119—128. [35] O’Brien,D.P.,Roazzi,A.,&R.Athias.An Experimental Investigation of Logical Reasoning by Tukano (Dase6)Speakers in Brazilian Amazonia and by North American University Students[Z].Manuscript under edi. torial review.201 1. [36] Over,D.E.&J.S.B.T.Evans.The probability of conditionals:the psychological evidence[J].Mind& Language,2003,(4):340—358. [37] Peacocke,C.A Theory ofConcepts[M].Cambridge,MA.:MIT Press,1992. [38] Quine,W.V.0. rd and Object[M].Cambridge,MA.:MIT Press,1960. [39] Rey,G.A not“merely empirical”argument for a language of thought[A].In Tomberlin,J.(ed.).Philo— sophical Perspectives 9 rC1.1995.201—222. [40]Rey,G.An Explanatory Budget for Connectionism and Eliminativism[A].In Horgan,T.&J.Tienson (eds.).Connectionism and the Philosoph) ofMind,Studies in Cognitive跏 啪drecht:Kluwer Academic Publishers,1991.219—240. (Volume 9)[C],Dor- [41]Rey,G.Contemporary Philosophy ofMind:A Contentiously Classical Approach[M].Oxford,UK:Basil Blackwell,1997. [42]Wittgenstein,L.The Philosophical Investigations[M].New York:Macmillan,1953. [43]Woods,W.Procedural semantics as a theory ofmeaning[A].In Joshi,A.,Webber,B.&I.Sag(eds.). Elements of Discourse Understanding[C].Cambridge,MA:Cambridge University Press,1981.300—334. 收稿日期:2012—06—04 作者简介:David P.O’Brien(1946一),男,美国纽约入,博士,教授。研究方向:认知心理学。 李淑静(1971一),女,河南开封人,博士,副教授。研究方向:认知语言学。 《现代外语》改版启事 由广东外语外贸大学主办、外国语言学及应用语言学研究中心承办的语言学期刊《现代外语》 将于2014年由季刊改为双月刊,继续遵循“学术性、前沿性、规范性”的办刊原则,意在更好地服务 于我国的语言学与应用语言学研究,以及外语学科的建设与发展。欢迎投稿!欢迎订阅1 41 

因篇幅问题不能全部显示,请点此查看更多更全内容

Top