Sentence Processing

There are two main levels of analysis in the comprehension of sentences. First, there is an analysis of the syntactical (grammatical) structure of each sentence; this is known technically as parsing. What exactly is grammar? It is concerned with the way in which words are combined. However, as Altmann (1997, p. 84) pointed out, "it [the way in which words are combined] is important, and has meaning, only insofar as both the speaker and the hearer (or the writer and the reader) share some common knowledge regarding the significance of one combination or another. This shared knowledge is grammar."

Second, there is an analysis of the meaning of the sentence. It is important to note that the intended meaning of a sentence may not be the same as its literal meaning. The study of intended meaning is known as pragmatics. Cases in which the literal meaning is not the intended meaning include rhetorical devices such as irony, sarcasm, and understatement. Some of the issues concerning pragmatics are discussed later in the chapter.

The relationship between syntactic and semantic analysis has been a matter of controversy. One possibility is that syntactic analysis generally precedes (and influences) semantic analysis; another possibility is that semantic analysis usually occurs prior to syntactic analysis; and a further possibility is that syntactic and semantic analysis occur at the same time. A final possibility is that syntax and semantics are very closely associated, and have more of a hand-in-glove relationship (Altmann, personal communication). These issues will be addressed shortly.

Grammar or syntax

An infinite number of sentences is possible in any language, but these sentences are nevertheless systematic and organised in various ways. Linguists such as Chomsky (1957, 1959) have produced rules to take account of the productivity and the regularity of language. A set of rules is commonly referred to as a grammar. Ideally, a grammar should be able to generate all the permissible sentences in a given language, while at the same time rejecting all the unacceptable ones. For example, as Harris (1990) pointed out, our knowledge of grammar allows us to be confident that "Matthew is likely to leave" is grammatically correct, whereas the similar sentence "Matthew is probable to leave" is not.


It might seem that parsing or assigning grammatical structure to sentences would be fairly easy. However, there are numerous sentences in the English language (e.g., "They are flying planes") that pose problems because their grammatical structure is ambiguous. Some sentences are syntactically ambiguous at the global level, in which case the whole sentence has two or more possible interpretations. For example, "They are cooking apples" is ambiguous because it may or may not mean that apples are being cooked. Other sentences are syntactially ambiguous at the local level, meaning that various interpretations are possible at some point during parsing.

Much research on parsing has focused on sentences that are ambiguous at the global or local level. Why is that the case? Parsing operations generally occur very rapidly, and this makes it hard to study the processes involved. In contrast, observing the problems encountered by readers struggling with ambiguous sentences can provide revealing information about parsing processes. It is conceivable that some of the processes used with ambiguous sentences differ from those used with unambiguous sentences, but there is no clear evidence of that.

There is a major distinction between serial and parallel theories of sentence processing (see Pickering, 1999). According to serial theories, one syntactic analysis of a sentence is selected initially. If this analysis proves unworkable at some point, then a second syntactic analysis is selected, and so on. In contrast, it is assumed within parallel theories that multiple syntactic analyses are all considered at the same time.

Another distinction, which is related to that between serial and parallel, is between modular and interactive: "Possible models range from a highly modular architecture—in which lexical access strictly precedes parsing, which in turn strictly precedes semantic processing.—to fully interactive models that claim there is a single process which combines lexical, syntactic, semantic, and world knowledge constraints without distinction" (Crocker, 1999, p. 216). As might be expected, serial theories are nearly always modular, whereas parallel theories are interactive.

There are numerous versions of each kind of theory. For example, as Pickering (1999, p. 127 pointed out, "Parallel models differ,. depending on how many analyses are maintained, what kind of ranking [of analyses] is employed, how long the different analyses are considered for, or whether parallelism is only employed under certain conditions." We will consider the most influential serial approach (the garden-path model) and the most influential parallel approach (the constraint-based theory of MacDonald, Pearlmutter, and Seidenberg, 1994). As you read about these theories, bear in mind that "The real issue.concerns which knowledge sources are used when" (Crocker, 1999, p. 219).

Garden-path model

Frazier and Rayner (1982) put forward a garden-path model, which was given that name because readers or listeners can be misled or "led up the garden path" by ambiguous sentences. The model was based on the following notions:

• Only one syntactical structure is initially considered for any sentence.

• Meaning is not involved in the selection of the initial syntactical structure.

• The simplest syntactical structure is chosen, making use of two general principles: minimal attachment and late closure.

• According to the principle of minimal attachment, the grammatical structure producing the fewest nodes (constituent parts of a sentence such as noun phrase and verb phrase) is preferred.

• The principle of late closure is that new words encountered in a sentence are attached to the current phrase or clause if this is grammatically permissible.

The principle of minimal attachment can be illustrated by the following example taken from Rayner and Pollatsek (1989). In the sentences, "The girl knew the answer by heart" and "The girl knew the answer was wrong", the minimal attachment principle leads to a grammatical structure in which "the answer" is regarded as the direct object of the verb "knew". This is appropriate for the first sentence, but not for the second. So far as the principle of late closure is concerned, Rayner and Pollatsek (1989) gave an example of a sentence in which use of this principle would lead to an inaccurate syntactical structure: "Since Jay always jogs a mile seems like a short distance". In this sentence, the principle leads "a mile" to be placed in the preceding phrase rather than at the start of the new phrase. In contrast, the principle of late closure produces the correct grammatical structure in a sentence such as: "Since Jay always jogs a mile this seems like a short distance to him".

There is increasing evidence that is hard to explain within the framework of the garden-path model. According to the garden-path model, prior context should not influence the initial parsing of an ambiguous sentence. However, there are several studies in which initial parsing did seem to be affected by context. For example, Altmann and Steedman (1988) recorded reading times to sentences such as the following:

1. The burglar blew open the safe with the dynamite and made off with the loot.

2. The burglar blew open the safe with the new lock and made off with the loot.

Altmann and Steedman presented a prior context sentence which referred to either one or two safes, and found that the nature of the context made a difference. When the context referred to two safes, this slowed down reading of "with the dynamite" in sentence (1). In contrast, when the context referred to only one safe, this disrupted reading of "with the new lock" in sentence (2). These findings suggest that context may influence initial parsing decisions.

Carreiras and Clifton (1993) studied readers' processing of sentences such as the following: "The spy shot the daughter of the colonel who was standing on the balcony". According to the principle of late closure, readers should interpret this as meaning that the colonel (rather than the daughter) was standing on the balcony. In fact, they did not strongly prefer either interpretation, which is contrary to the garden-path theory When an equivalent sentence was presented in Spanish, there was a clear preference for assuming that the daughter was standing on the balcony, which is also contrary to theoretical prediction.

Frazier and Clifton (1996) offered an explanation of these findings, but this explanation was "strikingly incompatible with the assumptions of the Garden-Path model" (Pickering, 1999, p. 137).


The garden-path model has various strengths. In particular, it provides a reasonably simple and coherent account of some of the key processes involved in sentence processing. However, there are various criticisms of the garden-path model, some of which are as follows.

First, it seems inefficient that readers and listeners should often construct incorrect grammatical structures for sentences. However, Frazier and Rayner (1982) claimed that the principles of minimal attachment and late closure are efficient because they minimise the pressure on short-term memory. They measured eye movements while participants read sentences such as those given earlier. Their crucial argument was as follows: if readers construct both (or all) possible syntactic structures, then there should be additional processing time at the point of disambiguation (e.g., "seems" in the first jogging sentence and "this" in the second jogging sentence). In contrast, according to the garden-path model, there should be increased processing time at the point of disambiguation only when the actual grammatical structure conflicts with the one produced by application of the principles of minimal attachment and late closure (e.g., the first jogging sentence). The eye-movement data consistently supported the predictions of the garden-path model.

Second, the assumption that meaning plays no part in the initial assignment of grammatical structure to a sentence seems implausible. For example, context studies (e.g., Altmann & Steedman, 1988) suggest that this assumption is incorrect. According to Pickering (1999, p. 140), "There is good evidence that semantic factors can have very rapid effects on ambiguity resolution. If a restricted account [e.g., garden-path model] is correct, then the initial stage that ignores these factors must be very brief indeed."

Third, it is unlikely that the initial choice of grammatical structure depends only on the principles of minimal attachment and late closure. For example, decisions about grammatical structure are influenced by punctuation when reading and by prosody (e.g., rhythm; stress) when listening to speech.

Constraint-based theory

An approach that has become influential in recent years is the constraint-based theory put forward by MacDonald et al. (1994). This theory is based on a connectionist architecture, and a key assumption is that all relevant sources of information or constraints are available immediately to the parser. Competing analyses of the current sentence are activated at the same time, with the analyses being ranked according to the strength of their activation. According to MacDonald et al. (1994, p. 685), "Ambiguity resolution is.a classic example of a constraint satisfaction problem. Multiple independent, partially redundant, probabilistic sources of information interact to allow the system to settle on an interpretation at each level."

According to the theory, the processing system makes use of four language characteristics in order to resolve ambiguities in sentences:

1. Grammatical knowledge constrains possible sentence interpretations.

2. The various forms of information associated with any given word are typically not independent of each other.

3. A word may be less ambiguous in some ways than in others (e.g., ambiguous for tense but not for grammatical category).

4. The various interpretations permissible according to grammatical rules generally differ considerably in frequency and probability on the basis of past experience.

Findings consistent with constraint-based theory were reported by Pickering and Traxler (1998). They presented their participants with sentences such as the following:

1. As the woman edited the magazine amused all the reporters.

2. As the woman sailed the magazine amused all the reporters.

These two sentences are identical syntactically, and both are likely to lead readers to identify the wrong syntactic structure initially. However, the semantic constraints favouring the wrong structure are greater in sentence (1) than in sentence (2). These constraints should make it harder for readers of sentence (1) to change their incorrect syntactic analysis when it needs to be abandoned (i.e., when the verb "amused" is reached). As predicted, eye-movement data indicated that eye fixations in the verb and post-verb regions were longer for those reading sentence (1), and there were more regressions.

Some of the findings discussed earlier are also consistent with constraint-based theory. For example, Altmann and Steedman (1988) found that context seemed to affect initial parsing. Constraint-based theory predicts context effects. Context produces activation of relevant information, and this then facilitates or disrupts processing of the following sentence.


A valuable feature of the constraint-based theory is the notion that there can be varying degrees of support for different syntactic interpretations of a sentence. As someone reads a sentence, the accumulating syntactic and semantic evidence gradually leads the reader to produce a definite syntactic interpretation. However, McKoon and Ratcliff (1998) pointed out that many of the details of syntactic processing are unspecified by MacDonald et al. (1994). For example, little is said about, "the problem of building syntactic structures for whole sentences when all the processor has to work with are the chunks of syntax stored in the lexicon for individual words. Until a simulation model can be developed to actually produce whole-sentence syntactic structures for a large variety and number of words and constructions, satisfactory evaluation of the constraint-based approach is not possible" (McKoon & Ratcliff, 1998, p. 35). Another problem was identified by Crocker (1999, p. 228): "Parallel models [such as constraint-based theory] predict that alternative, though dispreferred, structures are represented by the language processor. As yet, direct psychological evidence for the existence of such parallel representations has not been forthcoming."


As Harley (1995, p. 236) pointed out, "The study of pragmatics looks at how we deal with those aspects of language that go beyond the simple meaning of what we hear and say." Thus, pragmatics is concerned with the ways in which language is actually used for purposes of com munication. It is important to consider pragmatics, because the literal meaning of a sentence is often not the one that the writer or speaker intended to communicate. For example, it is usually safe to assume that someone who says "This weather is really great!", when it has been raining almost continuously for several days, actually thinks that the weather is terrible.

Austin (1976) argued that each sentence produced by a speaker has three rather different effects or forces:

1. The locutionary force: this is simply the sentence's literal meaning.

2. The illocutionary force: this is the speaker's goal in speaking, often called the intended meaning.

3. The perlocutionary force: this is the actual effect of the sentence on the listener.

According to this approach, listeners need to take account of the speaker's likely goals and the general context in order to understand what he or she is trying to communicate.

How do listeners or readers construct the intended meaning of a sentence? Clark and Lucy (1975) put forward a two-stage model, according to which the literal meaning is constructed before the intended meaning. They obtained support for this model by using sentences such as, 'I'll be very happy if you make the circle blue", and 'I'll be very sad unless you make the circle blue". These two sentences have the same intended meaning, but it took the participants much longer to understand the second sentence. According to Clark and Lucy, this occurred because the intended meaning of the second sentence differs from its literal meaning, and so both meanings had to be constructed.

Clark and Lucy (1975) used rather artificial sentences to support their two-stage model. More recent research has favoured a one-stage model, according to which people work out the intended meaning instead of (or at the same time as) the literal meaning. This model is supported by the common finding that intended meanings are formed as rapidly as literal meanings (see Taylor & Taylor, 1990, for a review). For example, Gibbs (1979) found that participants took no longer to process indirect requests (e.g., "Must you open the window?") than to understand literal meanings of the same sentences.

Gibbs (1983) obtained evidence that people can work out the intended meaning of a sentence without processing the literal meaning. Participants interpreted a sentence (e.g., "Can't you be friendly?") in terms of its intended meaning. After that, they were then presented with a similar sentence having a literal interpretation (e.g., "Are you unable to be friendly?"), and had to decide whether it was grammatically correct. The participants did not make this decision any faster than with control sentences, presumably because they had processed the literal meaning of the first sentence.

Cacciari and Glucksberg (1994) argued that much use of language is metaphorical, and involves intended rather than literal meanings (e.g., "My job is a jail"). We often interpret metaphorical language appropriately without being consciously aware of the fact that we are extracting the intended rather than the literal meaning. Why does this happen? Cacciari and Glucksberg (1994, p. 461) argued in favour of "the strong and still controversial claim that the comprehension and interpretive processes people use to understand language are common to literal and figurative [metaphorical] language use. Both types.. .require the application and integration of knowledge from linguistic and nonlinguistic domains."

Section summary

It has proved hard to distinguish between different theoretical accounts of ambiguity resolution. As Crocker (1999, p. 229) pointed out, "Each [model] has its own appeal,.. .and each can seemingly be made to account for empirical findings relatively well. Part of this stems from the fact that many models are only partially specified and implemented, if at all." Crocker (1999, p. 213) argued that serial models have certain advantages over parallel models: "Such models have much to recommend them: (1) they are conceptually simpler, (2) they are computationally simpler, in that less processing and memory resources are required, and (3)...they make strong, testable predictions about human behaviour."

An approach such as the garden-path model may have certain advantages, but it also seems to have severe limitations. In particular, MacDonald et al.'s (1994) theory can account for most of the findings more readily than can the garden-path model. As Pickering (1999, p. 145) concluded, "The emerging picture is of a system that integrates different sources of information on-line, in order that its choice of interpretation is based on all the sources of evidence that are available."

Comprehension for text as a function of text difficulty and opportunity to use subvocal speech. Adapted from Hardyck and Petrinovich (1970).

Stop Anxiety Attacks

Stop Anxiety Attacks

Here's How You Could End Anxiety and Panic Attacks For Good Prevent Anxiety in Your Golden Years Without Harmful Prescription Drugs. If You Give Me 15 minutes, I Will Show You a Breakthrough That Will Change The Way You Think About Anxiety and Panic Attacks Forever! If you are still suffering because your doctor can't help you, here's some great news...!

Get My Free Ebook

Post a comment