Guide to psycholinguistics pdf free download

The Routledge Handbook of Chinese Applied Linguistics is written for those wanting to acquire comprehensive knowledge of China, the diaspora and the Sino-sphere communities through Chinese language. It examines how Chinese language is used in different contexts, and how the use of Chinese language affects culture, society, expression of self and persuasion of others; as well as how neurophysiological aspects of language disorder affect how we function and how the advance of technology changes the way the Chinese language is used and perceived.

The Handbook concentrates on the cultural, societal and communicative characteristics of the Chinese language environment. Focusing on language use in action, in context and in vivo, this book intends to lay empirical grounds for collaboration and synergy among different fields. This volume is the first handbook dedicated to language attrition, the study of how a speaker's language may be affected by crosslinguistic interference and non-use.

The effects of language attrition can be felt in all aspects of language knowledge, processing, and production, and can offer unique insights into the mind of bilingual language users.

In this book, international experts in the field explore a comprehensive range of topics in language attrition, examining its theoretical implications, psycho- and neurolinguistic approaches, linguistic and extralinguistic factors, L2 attrition, and heritage languages.

The chapters summarize current research and draw on insights from related fields such as child language development, language contact, language change, pathological developments, and second language acquisition. Taking a transdisciplinary approach to research, the book builds upon recent theorizing and measurement principles from the fields of applied linguistics, cognitive science, psychology, psycholinguistics, psychometrics, educational measurement, and social psychology.

This book provides linguists with a clear, critical, and comprehensive overview of theoretical and experimental work on information structure. Leading researchers survey the main theories of information structure in syntax, phonology, and semantics as well as perspectives from psycholinguistics and other relevant fields.

Following the editors' introduction the book is divided into four parts. The first, on theories of and theoretical perspectives on information structure, includes chapters on focus, topic, and givenness. Part 2 covers a range of current issues in the field, including quantification, dislocation, and intonation, while Part 3 is concerned with experimental approaches to information structure, including language processing and acquisition.

The final part contains a series of linguistic case studies drawn from a wide variety of the world's language families. This volume will be the standard guide to current work in information structure and a major point of departure for future research. Bringing together experts from both historical linguistics and psychology, this volume addresses core factors in language change from the perspectives of both fields.

It explores the potential and limitations of such an interdisciplinary approach, covering the following factors: frequency, salience, chunking, priming, analogy, ambiguity and acquisition. Easily accessible, the book features chapters by psycholinguists presenting cutting edge research on core factors and processes and develops a model of how this may be involved in language change.

Each chapter is complemented with one or several case study in the history of the English language in which the psycholinguistic factor in question may be argued to have played a decisive role.

Thus, for the first time, a single volume provides a platform for an integrated exchange between psycholinguistics and historical linguistics on the question of how language changes over time. The Routledge Handbook of Applied Linguistics serves as an introduction and reference point to key areas in the field of applied linguistics. The five sections of the volume encompass a wide range of topics from a variety of perspectives: applied linguistics in action language learning, language education language, culture and identity perspectives on language in use descriptions of language for applied linguistics.

The forty-seven chapters connect knowledge about language to decision-making in the real world. The volume as a whole highlights the role of applied linguistics, which is to make insights drawn from language study relevant to such decision-making. The chapters are written by specialists from around the world. Each one provides an overview of the history of the topic, the main current issues and possible future trajectory.

Where appropriate, authors discuss the impact and use of new technology in the area. Suggestions for further reading are provided with every chapter. The Routledge Handbook of Applied Linguistics is an essential purchase for postgraduate students of applied linguistics. Those strengths are now enhanced by adding new chapters and thoroughly revising almost all other chapters, partly to reflect ways in which the field has changed in the intervening twenty years, in some places radically.

The result is a magnificent volume that can be used for many purposes. Aronoff and Rees-Miller have provided overviews of 29 subfields of linguistics, each written by one of the leading researchers in that subfield and each impressively crafted in both style and content.

I know of no finer resource for anyone who would wish to be better informed on recent developments in linguistics. Newmeyer, University of Washington, University of British Columbia and Simon Fraser University "Linguists, their students, colleagues, family, and friends: anyone interested in the latest findings from a wide array of linguistic subfields will welcome this second updated and expanded edition of The Handbook of Linguistics. Leading scholars provide highly accessible yet substantive introductions to their fields: it's an even more valuable resource than its predecessor.

New and thoroughly updated chapters by prominent scholars on each topic and subfield make this a unique, landmark publication. The editor's broad definition of the field ensures that the book may be read by those seeking a comprehensive introduction to the subject, but with little or no prior knowledge of the area. Building on the popular first edition, The Handbook of Linguistics, Second Edition features new and revised content reflecting advances within the discipline.

New chapters expand the already broad coverage of the Handbook to address and take account of key changes within the field in the intervening years. The book takes a three-pronged approach to examine what constitutes the phenomenon of the English language; why and in what contexts it is an important subject to study; and what the chief methodologies are that are used to study it.

In 30 chapters written by leading scholars from around the world, this Handbook covers and critically examines: English Language Studies as a discipline that is changing and evolving in response to local and global pressures; definitions of English, including world Englishes, contact Englishes, and historical and colonial perspectives; the relevance of English in areas such as teaching, politics and the media; analysis of English situated in wider linguistics contexts, including psycholinguistics, sociolinguistics and linguistic ethnography.

The Routledge Handbook of English Language Studies is essential reading for researchers and students working in fields related to the teaching and study of the English language in any context. This cutting-edge volume describes the implications of Cognitive Linguistics for the study of second language acquisition SLA. The first two sections identify theoretical and empirical strands of Cognitive Linguistics, presenting them as a coherent whole.

The third section discusses the relevance of Cognitive Linguistics to SLA and defines a research agenda linking these fields with implications for language instruction. Its comprehensive range and tutorial-style chapters make this handbook a valuable resource for students and researchers alike.

The Routledge Handbook of Applied Linguistics serves as an introduction and reference point to key areas in the field of applied linguistics. The five sections of the volume encompass a wide range of topics from a variety of perspectives: applied linguistics in action language learning, language education language, culture and identity perspectives on language in use descriptions of language for applied linguistics.

The forty-seven chapters connect knowledge about language to decision-making in the real world. The volume as a whole highlights the role of applied linguistics, which is to make insights drawn from language study relevant to such decision-making. The chapters are written by specialists from around the world. Each one provides an overview of the history of the topic, the main current issues and possible future trajectory.

Where appropriate, authors discuss the impact and use of new technology in the area. Suggestions for further reading are provided with every chapter. The Routledge Handbook of Applied Linguistics is an essential purchase for postgraduate students of applied linguistics. This book provides linguists with a clear, critical, and comprehensive overview of theoretical and experimental work on information structure.

Leading researchers survey the main theories of information structure in syntax, phonology, and semantics as well as perspectives from psycholinguistics and other relevant fields. Following the editors' introduction the book is divided into four parts.

The first, on theories of and theoretical perspectives on information structure, includes chapters on focus, topic, and givenness. Part 2 covers a range of current issues in the field, including quantification, dislocation, and intonation, while Part 3 is concerned with experimental approaches to information structure, including language processing and acquisition. The final part contains a series of linguistic case studies drawn from a wide variety of the world's language families.

This volume will be the standard guide to current work in information structure and a major point of departure for future research. The Oxford Handbook of Cognitive Science emphasizes the research and theory most central to modern cognitive science: computational theories of complex human cognition.

Additional facets of cognitive science are discussed in the handbook's introductory chapter. The Routledge Handbook of Chinese Applied Linguistics is written for those wanting to acquire comprehensive knowledge of China, the diaspora and the Sino-sphere communities through Chinese language.

It examines how Chinese language is used in different contexts, and how the use of Chinese language affects culture, society, expression of self and persuasion of others; as well as how neurophysiological aspects of language disorder affect how we function and how the advance of technology changes the way the Chinese language is used and perceived.

The Handbook concentrates on the cultural, societal and communicative characteristics of the Chinese language environment. Focusing on language use in action, in context and in vivo, this book intends to lay empirical grounds for collaboration and synergy among different fields.

Skip to content. Handbook of Psycholinguistics. Handbook of Psycholinguistics Book Review:. The Cambridge Handbook of Psycholinguistics. The Handbook of Psycholinguistics. Author : Eva M. The Handbook of Psycholinguistics Book Review:. The Oxford Handbook of Psycholinguistics. Author : M. You could have a study with two conditions: sentences that are sensible and sen- tences that are not.

Ideally, these sentences would differ as little as possible in every other way see the next section and so you might have a sentence set like: The boy rode his bike down the street and The boy rode his bike down the tree. You would also need to create and test at least 59 more sentence sets that differed in the same way. The easiest way to see this is when you are comparing two groups of people.

You would want to be careful to test two groups of people that are otherwise as similar as possible, because otherwise you would run the risk of having some other difference between the groups contribute to any difference in processing.

If there are differences between the groups, you cannot be sure whether it is due to their language background or their age. There are some statistical tricks you can do to deal with this, but it is easier and a better design to just make sure that age is held equal between the two groups to begin with.

These same concerns apply to materials in psycholinguistic studies as well—and in an analogous way. We are sampling a subset of individual people from a larger population, and we are also sampling individual language items from a large population of language.

It turns out that there are a number of properties of language that could, if ignored, cause problems for correctly determining the outcome of an experiment. It turns out that word length makes a contribution to certain measures of reading time. Other known factors include the frequency with which the word occurs in language use and the semantic relatedness between words. So, unless you are specifically interested in testing and manipu- lating , say, word length or frequency, you would want to make sure that your critical words did not differ in a systematic way between conditions with respect to these factors.

Sometimes genuine mistakes are made, but other times a research team may simply not take into account some potential difference that turns out to be impor- tant. Another team spots the difference and tests the hypothesis that it was this previously unknown factor that caused or con- tributed to the outcome.

Being Sneaky: Filler Items Another aspect of psycholinguistic experiments that is impor- tant, but not generally given too much airtime in descriptions of results, is the use of additional items that are not a part of the experimental design, but serve to distract participants from the true nature of the design or help ensure that they give valid responses. You might well use a lexical decision task, in which people are asked to decide whether a string of letters is a real word or not.

To appreciate why, imagine you were a participant in the study. In other cases, filler items are important to obscure the nature of what the researcher is interested in. Because you want your experimental items to be as similar as possible to avoid unintended effects, you have a set of materials in which the error, when it occurs, is always in the same place in the sentence.

So, you will probably need a lot of filler it is not unusual to see experiments with twice as many filler items as experimental items , and you will probably need to have some of those filler have errors in other parts of the sentence.

As a bonus, this also increases the likelihood that they will stay interested and alert while reading your sentences. Instead, they will often divide their sentences in groups and have participants see a subset of their materials.

For example, say you were still interested in the effects of grammatical errors. You make 20 sentence pairs, one of which might look like this: Grammatical: The key to the cabinets is on the desk. Ungrammatical: The key to the cabinets are on the desk. If you show both of these to the same participant, they will undoubtedly notice and start to get suspicious about why they are seeing two pretty much identical sentences, particularly if you show all 40 of the sentences 20 in each version so that every sentence that they see has a paired sentence that they see later.

So, you show the grammatical version of this sentence pair to half of the participants and the ungrammatical version to the other half. Between these two groups, you present half of the sentences in their grammatical version and other half in their ungrammatical version.

This is sometimes referred to as a repeated measures design and is certainly not unique to psycho- linguistics, although it is used quite often.

Incremental interpretation at verbs: Restricting the domain of subsequent reference. Cognition, 73, — Integration of word meaning and world knowledge in language comprehension. Science, , — Kutas, M. Psycholinguistics electrified II — Traxler Eds. New York: Elsevier Press. Menenti, L. When elephants fly: Differential sensitivity of right and left infe- rior frontal gyri to discourse and world knowledge. Journal of Cognitive Neuroscience, 21 12 , — Mitchell, D. On-line methods in language processing: Introduction and historical review.

Clifton, Jr. Brighton, UK: Psychology Press. Rayner, K. Eye movements in reading and information process- ing: 20 years of research. Psychological Bulletin, 3 , — Tanenhaus, M. Integration of visual and linguistic information in spoken language comprehension.

In fact, it is so often ambiguous that this bears repeating: Language is often ambigu- ous. What does this mean? Take the following sentence: 1. I went down to the bank yesterday. So, where did I go yesterday? Did I go to a place where money changes hands, or did I go to a place where water flows nearby? In particular, which meanings immedi- ately become activated? Context might help in this case, but the key question is, when?

As we saw in chapter 2, sentences have structure beyond the linear order of the words in them. And, it turns out that sometimes the same string of words can have more one possible structure. Take the following: 2. I watched the man with the binoculars. So, according to this sentence, who had the binoculars? The man? Both are possible. Finally, the thing that really makes sentences particularly ambiguous is the fact that sentences can have more than one possible syntactic structure temporarily, for some brief period of time during the sentence itself, but that by the end of the sentence only one structure is possible.

The florist sent the flowers was very pleased. Here, the sentence has one and only one syntactic struc- ture by the end—in which the florist is the subject and was very pleased is the predicate.

The question this kind of ambiguity poses is: Does the language processing system notice the temporary ambiguity?

If it does, how does the system handle it? Psycholinguists have spent a great deal of time and research energy on these types of ambiguities in language, but this is not because they are interested in ambiguities alone. The big question at the heart of this chapter, and in the research on ambiguity, is what kinds of information do people use to pro- cess language, and whether and how these different sources of information interact.

The Need for Speed Revisited In chapter 1, we discussed the need for speed in language processing. The speed required by language in fact creates a tradeoff between accuracy and speed—we need a system that is both fast and accurate, but speed may come at the cost of accuracy. On one side, the hypothesis has been that the best way to be as fast and accurate as possible is to have a system in which there are clear, limited information paths with multiple, separate processes that are specialized to handle each part of the lan- guage process individually.

This is the modular approach. On the other side, the hypothesis is that the system gains speed and accuracy by allowing multiple sources of information to interact as they are available—this is the interactionist approach. When the processing system allows information to flow in only one direction from one process to another, it is said to be a serial system.

When the system allows multiple processes to co-occur in time and for information from later processes to flow back and influence previous but still ongo- ing processes, then it is called a parallel system.

In this chapter, we will discuss the modular and interac- tionist approaches in detail as well as examine some of the experimental evidence in favor of each. To do this, we will focus on an area that has been at the center of the debate between these approaches: processing ambiguous words and sentences. This is a huge area of interest and a lot of factors appear to influence this process of recognition and meaning.

A summary of four major factors is given in Table 4. This is true for written language as well, for people who can read. A very famous example of this is the Stroop Test—in which people are given a list of words printed in different colors. You can try this yourself.

The Stroop Effect occurs because reading is automatic, but determining and naming col- ors is not. So, these two processes interfere with each other, and the identity of the words can affect how easy it is to say the color name. There are a few logical possibilities: Prior context could have no effect on initial access—this is the autonomous access model. The benefit of this model is that it can be fast. Or, prior context could constrain access in a way that allows only the context-ap- propriate meaning to be activated—this is the context-sensitive model.

The logic for this model is why waste energy accessing more than what you need given the context? So, all meanings are accessed, but the context-appropriate one is accessed earlier.

Initial investigation into this issue found evidence in favor of autonomous access in which context did not prevent alter- native meanings from becoming activated. For example, Onifer and Swinney used a cross-modal method in which par- ticipants listened to sentences that provided a biasing context toward one meaning of an ambiguous word or the other, and then, at that ambiguous word, they saw a different word on a computer screen that they needed to make a lexical decision to.

Onifer and Swinney found that words that were related to either meaning of the target word prompted faster response times than words that were semantically unrelated, regardless of the con- text that the target word was in. In critical trials, the probe word was always a real word and so what we are interested in is how quickly participants responded.

Crucially, they were faster to respond to fish compared to coal , even though the ambiguous word scale in this sentence very clearly refers to the kind of scale that weighs things and not something found on a fish. Further, the size of the increase in response speed for fish was basically the same as the increase in response speed for weight compared to source.

The fact that both fish and weight cause faster times supports the idea that in terms of information flow, sentence-level information is not initially used during the process of accessing a word. If it were, then we would expect to see benefits for context-appropriate meaning at the word itself—weight should show a larger increase in response speed than fish. These results are thus consistent with the auton- omous access model, sometimes also called exhaustive access because in this model all meanings of a word are accessed.

Interestingly, an important factor for ambiguity resolution appears to be the frequency of the different meanings of the ambiguous words. We can distinguish these two types of words by calling them balanced both meanings roughly equal and polarized one meaning more frequent than the other s. However, later experiments have found that mean- ing frequency is important, and many studies have found a subordinate-bias effect. This effect is as follows: in a neutral, non- biasing context, words that are balanced cause longer reading times than words that are either unbalanced or unambiguous.

What about in a context that biases toward the dominant or subordinate meaning before the word? In this case, there are longer reading times for the unbalanced word when the context is biased toward the subordinate meaning compared to context biased toward the dominant meaning or an unambiguous word.

This subordinate-bias effect has been studied quite a bit in order to determine which model best accounts for it and effects related to it.

Importantly, in terms of our models of meaning activation—a reordered access model would account for the subordinate bias effect by stating that the subordinate meaning may, with context, become as acces- sible as the dominant meaning, but the dominant meaning will still become activated, even when the context is strongly biased toward the subordinate meaning.

This means that frequency cannot be entirely overruled by context. In fact, both of these questions have posed problems for researchers: first, meaning dominance is usually determined by questionnaire studies that ask people to judge the meanings of words in various ways. Over the years there has been dis- agreement among researchers about materials, and part of the problem is that word use can be influenced by local conditions.

Biasing contexts are even more tricky. Martin, Vu, Kellas, and Metcalf provided data in which strongly biased con- texts cause the subordinate bias effect to go away, supporting the possibility of a selective model in which context may have a bigger role.

These authors argue in favor of a context-sensitive model of ambiguity resolution in which the strength of the bias in the context can overcome frequency bias, allowing context to have a larger influence when it is sufficiently biasing.

The custodian fixed the problem. She inserted the bulb into the empty socket. The gardener dug a hole. She inserted the bulb carefully into the soil.

These are examples of strongly biased contexts, with the ambiguous target word in bold. They compared these to weakly biasing contexts like the ones in the following: 6. The farmer saw the entrance. He reported the mine to the survey crew. The scout patrolled the area. He reported the mine to the commanding officer.

In one experiment, they added a second, naming task. In this case, a related or unrelated word would appear immediately after the participant had pressed the button after reading the tar- get, ambiguous word. This is evidence in favor of the idea that context does play a role in accessing the meaning of words. More recently, Sereno et al. It turns out that in the vast majority of stud- ies, response times to related words are compared to response times to these unrelated control words.

Also in most studies, each control word is matched in frequency to its corresponding ambiguous word. Remember that frequency plays an important role in processing words, with high frequency words processed more quickly, all else being equal.

However, the problem as Sereno et al. So, Sereno et al. They found that this made a difference. They used an eye-tracking methodology to examine reading times for sentences in which the context was biased toward the subordinate meaning of a polarized word.

First, they found a subordinate-bias effect for the target ambiguous word to the overall-frequency control word—meaning that the ambiguous word showed longer reading times than the control. This repli- cates and confirms previous work. However, when they com- pared the ambiguous word to a control word that was matched to the frequency of the subordinate meaning, then the oppo- site was true—the ambiguous word showed faster reading times than the control word.

It is not yet clear why this reverse effect happens, but this result highlights both the importance of minute details in experimental design and why we are still not certain to what extent context exerts an initial effect on mean- ing retrieval in word processing. Bottom-Line on Ambiguous Words: Context Counts One thing that experimental evidence does converge on is that context does influence how meanings of ambiguous words are accessed.

But, for our purposes here, the key, take-home mes- sage at this point is that despite early evidence against the role of sentence context in word meaning, it does appear that con- text can, at minimum, modulate the availability of appropriate word meanings. This time the question is—when and how does context influence how sen- tences are constructed instead of how meanings are accessed? Every word can have a grammatical category assigned to it- noun, verb, adjective, and so on.

As we discussed in chapter 1, different languages impose different rules about how gram- matical categories may be combined. These syntactic rules gen- erally apply without too much consideration for how sensible the meaning of a sentence is. So, we can have a sentence that is grammatically well-formed, but meaningless: 8. Swift tables never fly below three dogs in the afternoon.

And this is importantly different from a string of words that conveys a meaning, but is not grammatically well-formed: 9. Use to car need we go store to. We are reasonably sure, based on many experimen- tal results, that people process sentences in a largely incremen- tal fashion. This means that as we encounter each word we try to process it as fully as possible. From a working memory per- spective, this makes a lot of sense.

Upcoming, future words might provide disambiguating information that is inconsistent with how we initially build the sentence. Nonetheless, we seem to take the risk and interpret words and incorporate them into structure as we go. Key questions about how we do this include: Can context, particularly semantic and pragmatic information, influence how this happens? As with ambiguous words, ambig- uous sentences give us an effective tool to investigate this. If there are two or more!

What kinds of information influence structure building, and when? First, there are standing ambiguities: this is the case when there is no necessary reason to rule out or pick a given structure, and so the sentence has two or more possible structures even at the final word.

Sam watched the spy with the binoculars. Here, it is uncertain from this sentence alone who had the binoculars—Sam or the spy. In terms of structure, this sentence is ambiguous because the preposition phrase with the binocu- lars can be attached either as a phrase that modifies the spy or attached to the verb phrase itself, in which case it modifies how the watching was done.

Simplified syntactic structures for these two interpretations are given in Figure 4. These are harder to detect because they are, as the name suggests, tem- porary. They occur when two or more possible structures are possible in a sentence but that at some point one structure is unambiguously the correct one.

This point is called the disam- biguation point. Take for example: The horse raced past the barn fell. This is the canonical example sentence for temporary ambi- guities. Upon first encountering this sentence, many people find it difficult to get the correct interpretation of it, or even believe it is ungrammatical. The glass submerged in the water cracked.

Here, the interpretation of the sentence is a bit clearer— there is a glass that was submerged in water and this same glass cracked. We can now apply this to the horse in 11 : there is a horse that was raced past a barn, and this same horse fell down. Does context matter in how we process these sentences?

The very fact that 12 is easier than 11 is an important clue. A full version of these sentences would look like this: The horse that was raced past the barn fell. The glass that was submerged in the water cracked. Crucially, both the that and the was can be omitted in English—this creates the ambiguity by making the set of words for the relative clause temporarily indistinguishable from a sentence without a relative clause, at least for many verbs.

Example: The horse raced past the barn. We can end this sentence after barn because up to this point there is another syntactic structure that can be applied, in which raced is the main verb of the sentence, not part of a description of the horse, and so the horse is doing the racing, and then an additional phrase tells us where.

In fact, we could even poten- tially end the sentence after raced. So, back to the main questions: What do we do when we encounter these ambiguities? Do we entertain all possible interpretations or do we pick one and stick with it until forced to abandon it?

Can prior context help disambiguate, and if so—when does it have this effect? These questions are remark- ably similar to those posed for ambiguous word meanings, and the answers that have been proposed are also quite similar. In the case of structural ambiguity, there are two main mod- els that have proposed: The garden path model and the con- straint-based model. The garden path model is similar to the autonomous access model for word meaning—in this model prior semantic context has no initial effect.

Another big difference between these two models is that the garden- path model is a two-stage model while the constraint-based model is a one-stage model. This means that in the garden- path model processing happens in two distinct steps while in the constraint-based model everything happens as part of the same process. Being Led Down the Garden Path. As with the exhaustive access model for word processing, the garden path model achieves speed by simplifying the process: Context does not have an immediate impact on building sen- tence structure.

Temporary ambiguities like the one in 11 are often referred to as garden path sentences, as in the com- prehender is being led down and abandoned in the garden path to the wrong place.

In the garden path model e. Importantly, in the first stage, only one structure is constructed, even if others are possible. Then, if the structure is subsequently incompatible with further syntactic information, or semantic or pragmatic information, the structure is revised in the second stage. The crucial part of this model, and what really distinguishes it from other models, is this first stage.

If the words coming in came from more than one structure, but the processing system or parser only builds one structure, a good question is what structure does the parser build, particularly because according to the model, no semantic or pragmatic information is available at this point during processing.

The garden path model has two key principles that apply to how words are initially attached to the sentence structure: minimal attachment and late closure. But, using a simplified version of syntactic structure building, we can see how this principle can apply during processing. In this case, a different structure must be used. Figure 4. The prob- lem is that when readers encounter the beginning of the PP, by, it is unclear what to do with this phrase.

Notice here that in order for this interpretation to be correct, we must basically posit some unspoken words that are shown in parentheses, who and was, and build a bunch of extra struc- ture to accommodate the relative clause. The key idea here is that the second structure in Figure 4. Unfortunately for the comprehender in this case, the first structure, while simple and following minimal attachment, is not ultimately correct and so according to the garden path model, this initial structure is recognized as incorrect at the word by, and the sentence structure must be revised at this point to match the actual input.

The second principle of the garden path model is most clearly illustrated by standing ambiguities. This explains why there is a preference in English to interpret the binoculars as being with the spy in the sentence 10 , repeated as 16 , rather than Sam.

Recall from Figure 3. Of course, if there are pragmatic or semantic reasons why this would create a nonsen- sical sentence, as in the following sentence, then in the second stage of the model, the structure is revised to attach high to the verb phrase. The spy watched the bomb with the binoculars. Frazier argued that late closure is helpful from a processing perspective because it has a lower processing cost.

The problem with this particular principle is that it does not appear to be true for all languages. Consider the following two sentences from Cuetos and Mitchell The journalist interviewed the daughter of the colonel who had had the accident.

El periodista entrevisto a la hija del coronel que tuvo el accidente. Cuetos and Mitchell found while English speakers gener- ally preferred an interpretation of 18 in which it was the colo- nel and not the daughter who was involved in the accident, Spanish speakers preferred the opposite in a sentence like Further, eye-tracking data showed that the preference for high attachment in Spanish was present early in the sentence and so was not the result of a process in which late closure first applied and then was overridden.

This is not limited to Spanish or English —some lan- guages appear to take one approach while other languages take the other. Nonetheless, it could be argued that it is still the most efficient strategy for the parser, and the default. In this case, some lan- guages, for whatever reason, are exceptions to this processing default.

Another possibility, and one advocated by proponents of constraint-based models, is that what looks like a late clo- sure preference is simply a reflection of the statistical frequency of certain structures in the language. So, in English, the final interpretation of a sentence involving a low attachment is more frequent than an interpretation with high attachment, and so comprehenders come to expect and anticipate low attachment. Conversely, speakers of Spanish come to anticipate high attach- ment because that is the structure more frequently encountered in Spanish.

Instead of applying general principles regardless of con- text which sometimes fail to give the correct interpretation , many types of information are used to incorporate an incoming word into a sentence structure. In the case of ambiguous cases, all possibilities are considered and ranked according to their likelihood, again based not only on syntactic information, but things like the frequency of occurrence for a particular struc- ture and the semantic fit between the rest of the sentence and the current word.

Mistakes can still be made, particularly when evidence from these sources is pointing in one direction but the ultimate correct structure is something else. Why is it that 20 is so difficult while 21 is not? Raced occurs more frequently as a past tense form of race and so this may bias the processing system to rank the main verb interpretation higher than the past participle.

Also, there is the semantic fit between horse and raced. As an animate thing, it is possible for a horse to do things rather than having them only done to it , and further, horses are well known to race. Again, this causes the main verb interpretation to be ranked higher than the reduced relative. Compare this to 21 : sub- merged is also ambiguous between a past tense and past par- ticiple form. However, in this case the fit between glass and submerged, unlike horse and raced, means that it is unlikely that the glass submerged something because the glass is an inani- mate object and so in a literal context it is unlikely to be the thing doing the submerging.

This type of factor can cause the reduced relative interpretation of the sentence to be ranked higher than the main verb interpretation, or at least have the reduced relative alternative be more readily available when the disambiguating cracked is encountered. Another type of temporary ambiguity arises in English between whether a noun following a verb is a direct object or the beginning of a new sentence embedded within the main sentence.

For example: The student learned geometry over the summer. The student learned geometry was hard over the summer. In 22 , there is a student who is learning a thing— geometry.

In 23 there is a student who is learning a relation— geometry is hard. This will be true regardless of the verb. However, according to the constraint-based model, if a particular verb, such as confess or realize, does not frequently occur with a direct object, then this should influence the syntactic structure that is initially highly ranked or chosen. In particular, if a verb does not usu- ally occur with a direct object, then if a noun immediately fol- lows the verb it may well be interpreted as the beginning of an embedded sentence.

In short, according to the constraint-based model, the lexical information specific to particular verbs is used from the very beginning during sentence processing, while in the garden path model verb-specific information is only used in the second stage of processing.

Before continuing, an important reminder is in order. The comprehender him- or herself might not notice that one sentence is more difficult than another. It is important to emphasize this, because it makes the job of distinguishing the two models more difficult. We are looking for differences that are below the threshold of conscious attention and that occur in the very first moments of encountering each word in a sentence.

With this idea in mind, we can turn now to some of the experi- mental evidence regarding ambiguous sentences. Garden-Path or Constraint-Based? The Evidence Weighs in One of the reasons that this topic has lasted so long in psycho- linguistics, that is, why researchers could not agree about the flow of information in how sentences are processed, is that the evidence has been unclear, with both sides claiming strong sup- porting evidence. This highlights both the difficulty of design- ing experiments to test for differences in the very earliest stages of processing sentences, and the importance of testing what one thinks one is testing.

Part of the problem has been that because the models are really distinguished by how they initially deal with a word, it has been difficult to conclusively show that con- text has an effect on this earliest stage. Context clearly does have an effect, but in order to distinguish the models, it must have an effect on initial processing.

One way that this has been tested is by looking at what happens when the reader encoun- ters the word that disambiguates the sentence. The basic idea is that if this word disambiguates in favor of the initial interpreta- tion, then reading times for this word should be as fast as an equivalent unambiguous version of the sentence.

Rayner and Frazier found this type of effect in sentences like 24 and The criminal confessed that his sins harmed many people. The criminal confessed his sins harmed many people. In this case, 24 is the unambiguous control condition because it contains the word that which marks the beginning of the embedded sentence his sins harmed many people.

The sen- tence in 25 should have the same structure, but because it does not contain the word that, it is briefly ambiguous: his sins could either be the beginning of an embedded sentence which it is or the direct object of confessed. The principle of mini- mal attachment predicts that readers should initially build a sentence structure in which his sins is a direct object. Rayner and Frazier found evidence that this was indeed the case: readers slowed down at harmed in 25 compared to So, it is clear that readers can be led down the garden path to the wrong initial interpretation, the question now is—is this caused because autonomous processing principles are applied regardless of context or is it because context is sometimes misleading?