Sentence Comprehension (II)

7. Assessing comprehension
8. Theories of asyntactic comprehension
9. Current thinking

7. Assessing comprehension

Imagine what a sentence would have to be like that forced the comprehender to rely solely on the grammar and not be comprehendible from gestures, expression in the voice, general knowledge, and all the other cues that we pick up on in conversation. Here is a sentence that does just that:

(1) It was the boy that the girl kissed.

This sentence only has word order and grammatical words to indicate who was kissing whom. If you had to say who did the kissing, which one of the participants would you pick? This is a slightly stilted way of speaking and we do not use a sentence in this form very often, but I hope you can see that it is the girl who is the one kissing. It is this sort of sentence that the patients with comprehension deficits find difficult to understand. They may end up guessing, or going for the first one mentioned, or picking the more likely one based on their knowledge of the world. When they cannot work it out from the grammar, their scores are near 50% for chance (guessing on a two-choice judgment) or maybe even below chance (if using some strategy, such as picking the first participant each time). You yourself probably would find this next sentence, (2), easier than the one above, (1):

(2) It was the girl that kissed the boy.

It means the same thing as (1) but is easier to understand, isn't it? The patients we are going to talk about also find (2) easier than (1) and make many fewer errors on it. The trouble is, they could be working it out on the basis of left-to-right allocation of roles: the first person does the action to the second person. This would be a consciously applied strategy, though it does reflect a predominant feature of English. (3) is another difficult structure for patients with sentence-comprehension difficulties, while (4) is not, except for the most severely affected.

(3) The boy was kissed by the girl.
(4) The girl kissed the boy.

Sentence (3), like (2), is easy because the one on the left does it to the one on the right, which is the more frequent arrangement in English. In other words, many English sentences have a subject, verb and object, in that order, and the actor-action-victim in (4) are in the same order; so it is easy for the patient to interpret this structure. Sentence (3) is called a passive structure and has the reverse arrangement: the first one mentioned is the one acted upon. So English expectations will lead the listener astray, in the case of the passive, unless the comprehender can interpret from the grammar.

7.1 Structures used in assessing comprehension in aphasia

Sentence (6) is called the object cleft construction. Various ideas exist as to why it is harder than other structures, for all of us, to comprehend. The listener has to understand, from the order of the words, that the subject of the sentence is the girl, the second one mentioned, not the first participant, the boy. Then this structure has to be linked up to the roles of kisser and kissee to understand who kissed whom. Just by a simple change in word order, to that in (7), the subject of the sentence becomes the first participant, who is the kisser. Sentence (7) is easier to comprehend, perhaps because of something to do with this aspect that the subject and role of Agent (kisser, here) are filled by the same noun phrase. We do not yet know why there is an exaggerated effect of complexity for patients.

7.2 Tasks used in assessing asyntactic comprehension
Lots of reasons have been put forward as to why some patients find this structure of object cleft, and often the passive too (The boy was kissed by the girl), as well as others, hard to handle in a comprehension task. It is likely that there are as many reasons as patients! Not many patients are investigated in enough detail, on enough different tasks, to know why they have trouble with such sentences. A big obstacle to this research is that we have few if any tasks which directly reflect function of the parser interpreter. The ones that are used for college students (such as one called cross-modal priming) do not always work with patients. Often tasks are limited to decisions at the end of the sentence, whereas the processes involved in understanding have started as soon as a word is heard. The one you tried earlier, actor identification, is the kind where comprehension is tapped after the sentence. It is thought that this method allows the listener to introduce his or her own ideas on how to do the task. These ideas are likely to involve short-cuts that are easier than applying the grammar. As a consequence, performance may reflect the strategy the person employs, which may or may not be effective, and may not be what the person is normally doing in comprehending conversation in everyday life. It is better to use tasks that conceal the real measure of comprehension and better still to use ones that measure comprehension as it is taking place.

7.3 The profile of asyntactic comprehension
In light of all these problems, we will now consider the theories for why certain types of sentences are problematical for some aphasic patients. Many theories no longer have adherents, so we will cover just the ones that remain viable contenders. The disorder we have been talking about, asyntactic comprehension, can best be described with a profile of the performance that often occurs on a particular task, called sentence-picture matching. Although this task is problematical, for the reasons outlined above, it is still valuable as a screening instrument for this disorder.

A person is considered to have so-called asyntactic comprehension if, on the sentence-picture matching task, s/he can handle sentences like (5), but is poor or at-chance on sentences like (6):

(5) It was the girl that the boy chased.
(6) It was the apple that the boy ate.

Can you work out why (6) should be possible for these patients to get right? Think of some way in which you could guess the correct answer without using the grammar; that is, without using word order and the little (grammatical) words. By just noting the participants, boy and apple, and the action, ate that is, using the major lexical items you can guess that it must be the boy eating the apple, not the apple eating the boy. So it is possible to ignore the syntax and just use the meanings of the individual words (lexical items) and knowledge of what goes on in the world to guess the picture which matches this particular sentence (of which one ate the other).

[Back to top]

8. Theories of asyntactic comprehension

Before dealing with individual theories, we need to clarify the two types that they fall into in the main. These two types concern: (1) theories to do with processing; and (2) theories to do with the storage of knowledge.

Another matter to mention is that, when the renewed interest in sentence- comprehension difficulty in aphasia started in the 1970s, there was one type of aphasic patient that had everyone excited. The disorder these patients exhibited is called agrammatism. Because the vast majority of patients with agrammatism tend to have damage in the front part of the brain on the left side (frontal lobe), the disorder is linked to Broca's aphasia, originally thought to be from damage located there. Broca was a nineteenth century physician-anthropologist who proposed that an area at the base of the left frontal lobe in most people was the control center for speech. Although Broca's first patient had no speech beyond a recurrent utterance, and would more properly be said to have a global aphasia, with poor comprehension for language as well as no connected speech, the name, Broca's aphasia, has come to be linked to the halting, slow, limited output of agrammatism, which resembles a telegram in having a majority of major lexical items. More recently, it is thought that Broca's aphasia arises from more extensive damage, particularly around the Sylvian fissure, called the peri-Sylvian area.

The reason psycholinguists and aphasiologists were interested in these patients was that their comprehension difficulties might shed light on whether there was a module for syntax, separate from that for the meanings of words (lexical items). If it turned out that these patients, who understood the meanings of individual words well, had trouble comprehending the grammatical words and inflections that they could not produce in speech, then this parallelism in comprehension and production would be strong support for there being a separate module for handling syntax. So the earliest theories for asyntactic comprehension focused on agrammatic patients and tried to link their production deficits to their apparent difficulty with comprehending the same items that they could not produce.

Zurif is probably the main person responsible for the psycholinguistic interest in aphasia that started in the 1970s. It was he and his graduate students (e.g., Zurif, Caramazza & Myerson, 1972) who proposed the idea of parallelism between comprehension and production in use of the closed class vocabulary. It was also Zurif, along with his former student, Caramazza, who made the discovery about these patients' difficulty with sentences where the participants are both capable of the action. Before dealing with those theories on why word order is difficult, we need first to discuss the contribution of the closed class vocabulary hypotheses.

8.1 Closed class hypotheses
At first, in the 1970s and 1980s, the whole class of closed class items was implicated as a problem for agrammatic patients. An influential hypothesis at the time, was put forward by Bradley, a doctoral student under Zurif. They proposed that the grammatical words were stored in two places, including an additional lexicon that could be accessed rapidly. Speed of accessing is related to how frequently words are used in the language. Bradley, Garrett and Zurif (1980) argued that the normal rapid accessing procedures were lost and that only the slower route to the general lexicon holding all lexical items was available. This idea is no longer in favor, partly because it was not replicable (though some other supporting evidence Bradley produced has stood the test of time), but also because subsequent researchers showed that many patients are able to distinguish grammatical from ungrammatical sentences as efficiently as control subjects, solely on the basis of these words (e.g., Baum,1989; Shankweiler, Crain, Gorrell & Tuller, 1989).

Investigations into the closed-class vocabulary, from a comprehension point of view, have been prolific over the last quarter of a century. Certain studies in the early days, however, stood out as important. One such study is that by Linebarger, Schwartz and Saffran in 1983. Because it marks a turning point in the field, we will deal with it in detail, and give you a chance to try the task yourself.

Linebarger, Schwartz and Saffran (1983) showed that patients who failed on sentence-picture matching could nevertheless determine whether the following sentences were grammatical or ungrammatical:

I want you to go to the store now. I want you will go to the store now.
The boy put the book in the kitchen. The boy put the book that was in the kitchen.
Has the girl eaten her breakfast yet? Has the girl has eaten her breakfast yet?
Which animal did the girl want her mother to feed? Frank thought was going to get the job.
Which animal did the girl want her mother to feed? The hungry girl wanted her mother to feed.

As you can see, the ones on the right have had particular grammatical words substituted, omitted or added in order to make the sentence ungrammatical (there were other items in which word order was tampered with, but we will just address this set here). Thus, it seems that these patients, who could not match sentences to pictures with participants reversed, nevertheless must have possessed quite sophisticated grammatical knowledge in order to be able to judge whether the above items were grammatical. Firstly, grammaticality judgments tap awareness of grammar, but do not require interpretation. On the other hand, sentence-picture matching does require that the listener interpret sentence meaning, in order to be able to match the sentence to the correct picture. So these patients seemed to have quite a lot of knowledge, more than had been realized up until that time. Other laboratories have found the same thing.

Figure 1:
Four patients on good (gray bar), and poor (red and/or blue bars),
subsets of Grammaticality Judgments (Linebarger et al., 1983)

8.2 Parallelism Hypothesis
Another way in which the Linebarger, Schwartz and Saffran (1983) article was important was that not all their four subjects were Broca’s aphasic, yet all four had poor sentence-picture matching and good grammaticality judgments. Caramazza and Zurif (1976) also had included patients who were not Broca’s aphasic (they had another syndrome, called conduction aphasia, where the principal sign is difficulty with repetition), yet who produced similar performance profiles. Over the years, more and more evidence has accumulated on the range of expressive disorders that may be accompanied by performance lumped under the asyntactic banner. Thus, the parallelism proposal, expounded by Berndt and Caramazza (1980), among others, has disappeared from view in the writings of most researchers in the nineties and later. Apart from the Linebarger et al. study, what disposed of the parallelism hypothesis even more effectively was the finding of patients who had agrammatic output but no comprehension deficit, at least as far as assessment went (e.g., Nespoulous, Dordain, Perron, Ska, Bub, Caplan, Mehler & Lecours, 1988; Miceli, Mazzucchi, Menn & Goodglass, 1983).

8.3 Dissociations within the closed class
There is now abundant evidence that ayntactic comprehenders understand some closed class types better than others. Various people have tried to find some commonality in the types they can cope with, which would differentiate them from the types they handle poorly. Friederici, for example, long championed the idea that semantically rich functors were well handled by these patients whereas syntactically important ones were not. A semantic one would be the preposition by when it denotes location (e.g., The box was standing by the door), whereas a syntactic use would be by used with the passive, The girl was kissed by the boy. This difference does not always follow through, however, perhaps because different tasks tap awareness at different levels, or confound performance with task demands.

The same idea of semantic-syntactic dichotomy has been pursued in regard to the endings on words (inflections; e.g., raced). What confuses the picture is the fact that some patients do not discriminate ends of words in general too well (e.g., Tyler, 1992).

8.4 The link from loss theories to processing theories
Interestingly, asyntactic comprehenders can access the information contained in these grammatical markers when there is redundancy, such as several sources of information which converge. Yet when this same information is all that is available, they may not be able to use it. Bates’ and Smith’s teams have each done useful work to show this effect. A lot of their work is in languages more reliant than English on inflections. They also have shown that word order is more preserved in English agrammatic patients than the closed class, while the relative importance is reversed in those languages where inflections or functors carry the major grammatical information. Thus, there is general agreement that asyntactic comprehenders still have in long-term memory the grammatical information they need; but what is problematical for them is to access it or to use it as it arrives, in the normal flow of speech. The remaining theories to be covered will all be variations on this idea of a processing deficit to explain asyntactic comprehension.

8.5 Dissociations and double dissociations
Ever since the landmark study of Caramazza and Zurif in 1976, there has been awareness of the difficulty asyntactic comprehenders have with understanding sentences in which the action verb can be carried out by either participant, especially in structures where the predominant word order of English is not followed (called non-SVO, standing for non-Subject-Verb-Object word order). As you saw above, sentences in which either participant can do the action have now become the standard way of screening for asyntactic comprehension. Few studies do enough assessments to know the extent to which people with this word-order difficulty also have a vulnerability in understanding the closed class. It is known in at least a few patients that they do only have one or the other to a marked extent. This aspect of some patients’ having difficulty with one sort of information and not with another sort is called a dissociation. If two patients have the reverse pattern to each other, then we call it a double dissociation. These terms do not apply to just any difference, but certain criteria need to apply: for example, the difference needs to be quite marked (e.g., chance versus near perfect; or better than chance and worse than chance); efforts should also have been expended to rule out confounds and leave just a difference on a single variable that is important to information processing.

8.6 Verb deficits and sentence comprehension
One of the most studied dissociations in aphasia has relevance to asyntactic comprehension. This dissociation is between the production of verbs and that of nouns. It was originally intriguing because the dissociation seemed to be related to the site of lesion/type of aphasia. In other words, Broca’s aphasic patients seemed more often to have a problem with verbs, whereas Wernicke’s aphasic patients appeared vulnerable on nouns. We now know that many different kinds of patients have a verb deficit, including patients whose major difficulty is with nouns. Thus, it does not necessarily follow that a verb-production deficit leads to a sentence-comprehension impairment. In addition, there can be a deficit which seems to affect meaning in some people but the word forms (phonology) in others. The deficit itself differs in what tasks it affects (picture naming versus story telling) and different types of verbs may be more vulnerable than others. It is a problem to control confounds such as the frequency and imageability of verbs relative to nouns, but verb deficits still remain where people have made an effort to control these aspects.

Why a verb deficit is interesting in connection with asyntactic comprehension is because patients’ difficulties working out who did the action to whom in a sentence may be due to a problem accessing such information about particular verbs. A few group studies, however (Breedin & Martin, 1996; Berndt et al., 1997), have shown that not all asyntactic patients have trouble accessing verb information per se. There are sufficient case studies available now to realize that accessing information in the verb’s lexical entry may affect the ability to use that information to comprehend sentences in some patients, but not with everyone. It seems that there may be a whole host of reasons for patients’ failure to handle interpretation of sentences, only one of which is the accessibility or utility of information about the roles of participants an action verb takes.

8.7 The mapping hypothesis
Closely related to this idea of a verb deficit’s underlying asyntactic comprehension is the explanation Linebarger and her colleagues (1983) gave for good grammaticality judgments in patients with poor sentence-picture matching. Recall that this task dissociation was thought to be due to the interpretive demands of sentence-picture matching which were not required for detecting an ungrammaticality. What Linebarger, Schwartz and Saffran (1983) proposed was that these patients may not be able to work out, after grammatically processing a sentence, which participant in the verb’s action goes with the subject and which with the object (called mapping), especially when the agent is not the first one mentioned. They proposed, further, that either this mapping process itself is faulty (though they have given no details), or else the person has insufficient working memory resources to do both parsing and interpretation: parsing takes place (hence good grammaticality judgments) but interpretation exceeds resources (hence poor sentence-picture matching).

8.8 Capacity-limitation hypotheses

8.8.1 The trade-off hypothesis
This second alternative is called the Trade-off Hypothesis and introduces us to another group of theories which have some sort of capacity limitation in working memory. The trouble with this idea is that some of the resources required to do sentence-picture matching seem to be non-linguistic. Non-linguistic working memory demands were discussed earlier, when we introduced the sentence-picture matching task. So, patients could fail on this task, but not from a limitation in linguistic working memory resources, but rather from the excessive demands of the task itself, outside language (in the cognitive processes required to make their choice between the matching and non-matching picture) as demonstrated by Cupples and Inglis (1993).

Since then, Saffran, Schwartz and Linebarger (1998) have favored a model of comprehension akin to the constraint-based theories, in which syntactic and semantic (i.e., roles of participants in the action) information is processed interactively. In their current view of the asyntactic disorder, weak exploitation of syntactic information is overridden by lexical information, so that the person tends to put together the participants in the action as they exist in the real world. The task they and others have tended to use, instead of sentence-picture matching or actor identification, is so-called plausibility judgments. Here, the action between two participants is plausible in one order but implausible when the participants are reversed, as in (7) and (8):

(7) The man was carried by the horse.
(8) The horse was carried by the man.

The response required is Yes (for sensible) for (7) and No (for silly) for (8). Unless the comprehender can use the word order and the grammatical words like was and by, s/he will find this decision difficult to make, especially when the structure has a non-SVO word order.

8.8.2 Working memory and limited capacity
Another working memory hypothesis was proposed by Just, Carpenter and Miyake, appearing in various articles in the early 1990s. In their view, there is one working memory capacity for all linguistic tasks, including holding lists of words briefly. Furthermore, this all-embracing capacity is different in different individuals, and is reduced in brain damage. In asyntactic patients, the capacity limitation is supposed to have more impact on complex structures than on simple ones. The big weakness in this hypothesis is that it is only one possible account, which is hard to falsify, yet which explains the data no better than others (see Martin, 1995, for a critique).

8.8.3 Working memory for language
Yet another working memory hypothesis comes from Caplan and Waters (1999). They expand on Miyake et al.’s (1994) model in the number of working memory systems, by having one for language and another for what they call post-linguistic processes; in other words, what we earlier described as the nonlinguistic cognitive demands of comprehension tasks. This post-interpretive capacity is the same memory underpinning span tasks, in their view. So, in that respect, it differs from Cupples and Inglis’ (1993) conception, which is not shared by span tasks (their two cases had normal or borderline-normal spans). The problem for Caplan and Water’s model is that STM-deficit patients should not be able to handle the plausibility judgments involving misordered items, where there is a silly meaning in the syntax but a sensible meaning from the logical combination of participants), if span and post-linguistic functions rely on the same memory system; yet there are STM patients who do.

8.8.4 Working memory and multiple capacities
If you are not totally confused by the number of different working memory hypotheses around, there is one final one we need to consider: that from Martin’s lab (Martin & Romani, 1994; Martin & He, in press). This is the most elaborate yet in that it has a memory buffer for each particular code involved in language and list recall. Thus, there would be one to handle phonological information (to do with speech sounds), another to handle semantic information (word meaning and semantic roles), presumably one to handle syntax, and so on. When brain damage occurs, it would be conceivable to find patients where these different buffers are differentially damaged, leading to dissociations. Accordingly, Martin and Romani (1994) and Martin and He (in press) describe two patients who provide data conforming to the double dissociation pattern: ML who has a semantic STM deficit and EA who has the phonological STM pattern. How these different memory deficits have an impact on comprehension is governed by the type of information involved. If the sentence has a heavy semantic load, such as strings of adjectives which must be held in memory before the noun that they modify for example, the shiny, new, blue SAAB, then the semantic STM patient will have difficulty, but not when the same adjectives are re-arranged to follow the noun: The SAAB he drove was shiny, blue and new. Phonological STM is thought by most people in the field now, not to be required for parsing or interpretation in real time. That is because the information being processed, as we saw earlier, is no longer phonological. Hence, a phonological STM patient typically copes with sentence-comprehension tasks.

Where phonological STM may become important is if the patient also has some temporal processing disorder (Martin, 1993). What Martin (1993) suggests is that such a patient may need to hold onto the phonological input longer than normal if his/her language processes are operating more slowly since the stroke. This hypothesis has yet to be put to an empirical test, but it deserves attention.

8.9 Slow access versus rapid decay
The most likely candidate for a temporal processing disorder is the asyntactic comprehender. A number of different research labs have been pursuing that idea, ever since Bradley, Garrett and Zurif’s (1980) important study. The exact nature of the problem is still not clear, and it seems possible that different problems may afflict different people. Two ways in which people have thought temporal processing may be aberrant is: (1) slowed access to stored information, and (2) rapid decay of activation after stored information has been accessed (Kolk, 1995; Haarmann & Kolk, 1991). While there is evidence for both of these patterns, there is quite a lot of inconsistency between laboratories, so that the story has yet to unfold as to what the range of deficits is, and whether the variety of deficits explains corresponding difficulties with sentence comprehension. It is noteworthy that not just Broca’s aphasic patients show the patterns (see Tyler, Ostrin, Cooke & Moss, 1995). While Kolk does not believe that these two patterns will likely exist in the one patient, the first seems to be a problem with accessing, something that is well recognized in other, more easily tested areas, such as lexical semantics or the acquired dyslexias, while the second has to do with decay in memory. While the same piece of information may either not be accessible or may quickly lose its activation after access, it would be necessary to explain how damage produces one deficit and not the other.

[Back to top]

9. Current thinking
One of the problems in researching this topic is the difficulty in tapping real-time language processing. A few labs (especially Swinney’s) have advocated cross-modal tasks. In these paradigms, the research participant has to make a decision regarding a written word (such as whether it is a real word in the language lexical decision) appearing on the screen at some specific point during the presentation of an auditory sentence. The idea is to look for a fast reaction to the written word when it is related to some other word in the auditory sentence. More than one laboratory has found that aphasic patients are not able to respond reliably on this task (e.g., Tyler’s and Martin’s). While alternative tasks exist (e.g., word monitoring but it is only appropriate for parsing), other options are to use modern sophisticated equipment, such as ERP (event-related potentials a form of EEG), fMRI (functional magnetic resonance imaging a non-invasive scan of the brain), or PET (positron-emission tomography an invasive scan of the brain using a radioactive substance). In addition to being more or less invasive, these techniques vary in whether they better assess localization of function (fMRI, PET) or temporal processing (ERP). This last technique of ERP is the best equipped to tease apart the temporal aspects of language processing, but it has not yet been used extensively with aphasic patients.

Irrespective of the difficulty of tapping real-time language processing, the research done so far seems to indicate that the entity called asyntactic comprehension is far from "asyntactic" and is not likely to be due to a single etiology. In view of the variability in performance profiles and varying degrees of difficulty with different structures, there is a need for comprehensive single-case studies using a range of tasks. Syndrome labels like this one are no more than a useful shorthand to convey some idea of a patient's presenting profile; they convey little useful information when it comes to understanding what processes have been affected in any particular individual.

Another characteristic of recent research into sentence comprehension is that patients with etiologies other than stroke have become the focus of attention. In particular, Alzheimer’s sufferers have been of interest because of their relative preservation of syntactic processing compared to lexical semantics. It seems that, in testing them on the range of sentences from aphasic asyntactic comprehension research, these patients have trouble with the non-linguistic working memory demands of tasks and not with linguistic parsing and interpretation per se.

The research on patients with sentence comprehension difficulties is only now beginning to bear fruit in terms of explaining aspects of normal processing of language. The early hope that agrammatic patients' perceived difficulty with comprehension would indicate the existence of a syntactic processor common to both comprehension and production has not survived. The reason for failure may just be the lack of sensitive enough assessment tools or the rarity of the right patients to demonstrate a damaged central parser. On the other hand, Martin's demonstration of an effect of semantic information on comprehension is a breakthrough in showing a link between memory and language.

[Back to top]