Processing Distinctions Between Stems and Affixes: Evidence from a Non-Fluent Aphasic Patient*

cognition, 36 (1990) 129-153

Processing distinctions between stems and affixes: Evidence from a non-fluent aphasic patient*

LORRAINE K. NLER SUSANBEHRENS HOWARD COBB University of Cambridge

WlLLlAM MARSLEN-WILSON MRC Applied Psychology Unit, Cambridge Received June 2, 1989, final revision accepted February 19, 1990

Abstract Tyler, L.K., Behrens, S., Cobb, H., and Marsten-Witson, W., 1990. Processing distinctions between stems and affixes: Evidence from a non-fluent aphasic patient. Cognition, 36: 129-153. In this paper, we study the ability of a non-fluent aphasic patient, BX, to comprehend morphologically complex words when they appear in utterance contexts. Wefirst establish that he is insensitive to the contextual appropriateness of both derived and inflected words. In a further experiment we show that he has no difficulty processing the stems of complex words and conclude th-zt his problem is with the bound morphemes themselves. We then ask whether this problem is due to his inability to access either the phonological form of a morphologically complex word or its semantic andlor syntactic content. We find that only the access of semantic and syntactic content is impaired. We conclude from these six studies that: (a) BN presents a counter-example to the claim that non-fluent patients have particular dificulty with those aspects of morphology which have a syntactic function; (b) BN processes both derived and inflected words by mapping the sensory input onto the entire full-form of a complex word, but the semantic and syntactic content of the stem alone is accessed and integrated into the context. The semantic and syntactic implications of the suffix are never evaluated. This implies separate representation of the stems and sufSixes of some types of morphologicully complex words.

*We thank Paul Warren for his help with these experiments. This research was supported by an MRC ?ragramme grant to L.K.T. and W.M.W. Requests for reprints should be addressed to Lorraine K. Tyler. Department of Experimental Psychology. University of Cambridge, Downing Street, Cambridge CB2 3EB. U.K.

0010-0277/90/$8.00 O 1990-Elsevier Science Publishers B.V. 130 L. K. Tylar at al.

Introduction There are a number of reports in the literature of aphasic patients who have particular problems in producing morphologically complex words (e.g., De- Villiers, 1978; Goodglass & Berko, 1960; Goodglass, Gleason, Ackeman, & Hyde, 1972; Miceli & Caramazza, 1987). English-speaking patients of this type rarely spontaneously produce morphologically complex words and also tend to omit the free-standing grammatical morphemes (such as "the" "and" "but"). Furthermore, they are less likely to produce inflected than derived words. Problems with morphologically complex \vords show up even more clearl;. in the spontaneous speech of patients who are native speakers of languages like Italian and Hebrew, which have a richer inflectional system than English. Although Italian and Hebrew patients produce inflected words (presumably because an uninflected root is a non-word in these languages), the inflections tend to be inappropriate for the context (Grodzinsky, 1984; Miceli & Caramazza, 1987). There are indications that the same kind of difficulty with morphoZogically complex words also occurs in comprehension, although the data here are much more sparse (e.g., Eling, 1986; Patterson, 1979; Tyler & Cobtb, 1987). Although most of this research does not distinguish inflected from derived words, the studies that do (e.g., Tyler & Cobb, 1987) reveal the same kind of poor performance for inflected words in comprehension as in production. Although it is rarely the case that such patients only have problems with the closed class morphology, it has been claimed that difficulties with this particular type of linguistic element is at the heart of their language disorder because of the syntactic role which the closed class morphology plays in many languages. Free-standing grammatical morphemes and the inflectional inor- phology, in particular, contribute to the construction of a syntactic representation of an utterance (e.g., Anderson, 1982). The ciaim is that agammatic patients have problems with these linguistic elements because they do not have "knowledge of the structural roles played by grammatical markers ... (and) ... they are thereby unable to use these items as markers of phrasal constituents" (Cooper & Zurif, 1383). This pattern of impaired performance is consistent with models of the unimpaired processing system, such as Garrett's production model (1980), or Miceli and Caramazza's (1987) m~delof the structure of the lexicon, which stress the functional relationship between the closed class morphology and the syntactic structure of an utterance. In Garrett's model, for example, function words and grammatical morphemes are inserted into the sentence frame at the positional level so as to define the grammatical structure of an utterance. This accounts for the frequently observed CO-occurrenceof syntac- Processing distinctions between srems and ames 131

tic and morphological deficits in "agrammatic" patients (but see Berndt & Caramazza, 1980, Goodglass & Men, 1985; Martin, Blossom-Stach, & Feher, 1989). Errors in this stage of the process result in two types of errors involving affixes - omissions and substitutions -which are the types of errors found in "agrammatic* speech. For production deficits involving closed class morphemes, theory and data neatly converge. However, the situation is not quite so clear-cut when we consider comprehension deficits. There is little evidence bearing directly on the claim that when patients have problems in comprehension which involve morphologically complex words, it is because they have a general syntactic deficit. This is because there are no studies (with the exception of Lukatela, Crain, & Shankweiler, 1988; Tyler & Cobb, 1987) testing the comprehension of bound grammatical morphemes in sentence contexts. Instead, experiments typically involve presenting morphologically complex words in isolation - usually for the patient to read (Eling, 1986; Patterson, 1979). The only work which has examined the processing of grammatical markers in context has focused on the free-standing morphemes (e.g., Goodenough, Zurif, & Wein- traub, 1977; Grossman, Carey, Zurif, & Diller, 1986). This neglect is surpris- ing, since the only way to determine whether a patient can exploit the syntactic (or semantic) properties of a bound morpheme - whether an inflected or derived form - is to place it in a context which allows the functional signifi- cance of those properties to have an effect. This means that the morphologically complex word should appear in a sentential context rather than in an unstructured word list. However, to determine whether a patient has problems comprehending morphologically complex words, we have to do more than just look at his or her difficulty in processing such words when they appear in sentential contexts. We have to try to tease apart the possible underlying causes of the problem. This can be complex because there are a number of ways in which the piocess of comprehending a derived or inflected word can be disrupted. We assume here a three-stage process. The first stage involves mapping the senwry input onto mental representations of the plrcse!ogca! farm of a worct. In the absence of any data to the contrary, we assume that this process is the same for both simple and suffixed words - that is, the sensory input is mapped onto the entire form of the word. The second stage occurs after these form representations have been activated. This is when the syntactic and semantic properties of the word (i.e., its "lexical content") are accessed. The process of accessing this information may differ, depending on whether the meaning of the word represented in a decoinposed or whole form in the mental lexicon (e.g., Butterworth, 1983; Stemberger & McWhinney, 1986). The third stage is when the semantic and syntactic properties of a morpholog- 132 L.K. Tyler et al.

ically complex word are used in parsing and interpretation. Although it is possible in principle to distinguish between the access of content and its use in sentence interpretation (stages two and three), in practice this is very difficult and the experiments reported in this paper do not do so. In principle, patients may have problems with any of these stages. First, they may have difficulty mapping the sensory input onto mental representations of lexical form. This may be a general problem, involving both morphologically simple and complex words, or it may be confined to complex words. Second, they may have no problems with form mapping but the access of lexicd content may be impaired. They may be unable to process the syntactic andlor semantic proporties of morphologically complex words - either because they are unable to access the component morphemes, or combine them appropriately. Third, although they uay be able to access the content of morphemes, they may be unable to use them for parsing and interpretation. To further complicate the picture, deficits in the ability to access either lexical form or lexical content may bd confined to either derived or inflected words. To try to separate out these various factors, we carried out a series of experiments on a non-fluent aphasic patient, BN. These studies focused on different aspects of the invofved in comprehending morphologically comp!ex words. The first study was designed as a general test of BN's ability to inflected and derived words ihen they appear in sentential contexts. Since he performed poorly in this experiment, we then asked whether this was due to his inability to access the phonological form of such words, oi to his inability to access their semantic andor syntactic content. To answer these questions, we designed a series of studies in which morphologically complex words were either heard in an unstructured list of words or appeared in different kinds of sentential contexts. We designed the experiments so as to be able to tease apart BN's ability to process inflectional and derivational affixes as opposed to either the full-form of an affix-bearing word or its stem. We will first of all present a clinical profile of BN, including the results of standard neuropsychological tests. Then we will report data from a series of comprehension tests which focus specifically on his ability to process morphologically complex words.

The patient BN is a 44-year-old male native speaker of British English (born in 1944) who, in 1984, suffered a left hemisphere cerebrovascular accident (CVA). He is right-hand dominant with no appreciable hearing loss (mean hearing Processing distinctions between stem and afFrs 133

loss of 6 dB). No brain scan information is available for him. RN has a right side weakness and a non-fluent aphasia, as determined by assessment on the Boston Diagnostic Aphasia Examination (BDAE; Gooddass & Kaplan, 1972). Subtests of production abilities on the BDAE show LN's speech to be impaired in fluency, with melody, phrase length and gramma'ical form all below normal levels. We first tested BN one year after his stroke. His speech output was effort- ful and primarily consisted of an unstructured string of short noun phrases and very few verbs. Most of the words he produced were morphologically simple. He rarely produced an affixed word - whether inflected, derived, suffixed or prefixed. A sample of his speech is given below.

BN (describing cookie theft picture from BDAE): - Cookie ... boy ... eh yes and girl ... yes ... school ... eh ... cold ... taps ... plate ... and saucer. His memory capacity, as measured by digit span and digit matching tasks (where two separate digit strings were read to him and he was asked to indicate whether they were the same or different) was slightly below normal. He could only accurately recall or match up to four digits. BN was tested on various standard tests to determine whether his comprehension was normal. On the auditory comprehension sub-tests of the BDAE, his performance suggested that he only had a slight cornprehension deficit. He made no errors on the word discrimination and body prt identification sub-tests, and on the commands and complex ideational material sub- tests he scored with 75% accuracy. On the Token Test (De Renzi & Faglioni, 1978). his performance was severely impaired (31% correct). Finally, when he was tested on the Test for the Reception of Grammar (TROG; Bishop, 1982), which is a sentencetword-picture matching task, he made no errors in matching a spoken word to a picture. He was also very accurate on the sentence materials - including passives. The results of these auditory comprehension tests suggest that BN has some prob!ems in comprehending spoken language, but they do not specify the precise nature of the problem. We now describe the experiments we designed to test his comprehension of morphologically complex words.

Experiment 1: Derived and inflected words in context

This study was designed as a general test of BN's ability to process inflected and derived words in sentence. contexts (for a more detailed description of the materials and design see Tyler & Cobb, 1987). We had BN listen to pairs 134 L. K. Tyler et al.

of sentences where the second sentence of each pair contained a derived or inflected word which was immediately followed by a target noun (the word "cook in Table 1, which gives an example set of stimuli). For the test words, there was always a phonologically transparent relationship between the stem and the affix - that is, the affix did not cause phonological changes to the stem. The test word (italicised in the examples below) was either appropriate for the prior context. In the case of the contextually inappropriate test word, the sentence up to and including the stem was always appropriate. The form of the suffix determined whether the word-form as a whole was appropriate or inappropriate for the context. We established the appropriateness or inap ~ronriatenessof the test word bv means of h re-tests (see Tvler & Cobb. 1987, #or hetails). The frequency of he appropAste and inapp;opriate test.words were matched as closelv as ~ossible.The median freauencv of the sets of appropriate and inappropriate inflections were 14 andil, r~spectively,and the median freauencv of the aooro~riateand inaoorooriate.. . derivations were 11 and 6, resp&tive$ (~rancis& ~ucera,1982).

Table 1. E.~perimenrl: Example stimuli ------Derivations

1. Appropriate: Sarah could not understand why John used so much butter. He was the most wasteful cook she had ever met. 2. Inappmpriate: Sarah could not understand why John used so much butter. He was the most wasteage cook she had ever met. 3. Non-word: Sarah could not understand why John used so much butter. He was the most wasrely cook she had ever met.

1. Appropriate: I have to be careful when eating ice-cream. It often causes pain in my loose filling. 2. Inappmpriate: I have to be careful when eating ice-cream. It often causing pia in my loose filling. 3. Non-Word: I have to be careful when eating ice-cream. It often cadypain in my loose filling. -- Processing distinctions between stems and affixes 135

In a final condition, we had a test word which wzs a non-word comprised of a real word stem and suffix (always a derivation), but illegally combined (e.g., mixly, washness). This condition was included in case BN showed no difference between the contextually appropriate and inappropriate words. There were 21 derived and 24 inflected test words. These were pseudoran- domly interspersed with 43 filler items designed to obscure the regularities of the test items. We constructed three lists of materials from these test and filler items. Each list contained an equal number of items in each of the appropriate, inappropriate and non-word conditions, as well as the total set of fillers. Six practise items preceded the test materials. The three versions were recorded by a female native speaker of English at a normal conversa- tional rate. We then placed timing pulses at the onset of each target word. These were located on the non-speech channel of the tape and could not be heard by the listener. They functioned to initiate a digital timer which was stopped when the subject pressed a response button Because we were interested in BN's ability to integrate the syntactic and semantic implications of the suffix into the representation of the sentence as it is being constructed, we used the word-monitoring task. In this task, the subject hears a spoken sentence containing a target word, which is specified in advance, and has to press a response key as soon as the word is identified. Reaction times (RTs) to respond to the target word are measured from word onset. In our materials, the target word immediately followed the inflected or derived test word. RTs in the appropriate condition constitute the baseli~2 condition against which we can evaluate RTs in the other two conditions. ir' th= syntactic and semantic implications of morphologically complex words are immediately integrated into the sentence context, then RTs to a target word which follows a cxntextually appropriate suffixed word should be faster than those following contextually inappropriate test words. To the extent that this effect differs for inflected and derived words, then the difference between RTs in the appropriate and inappropriate conditions will vary. Apart from BN, we tested a group of 23 subjects (ranging in age from 20 to 37 years). Each of the control subjects was tested or-,sne ver^J:nn of the materials, whereas BN was tested on all three, with each testkg session separated by an interval of one month. The w.itrol subject< show an im- mediate effect of the contextual appropriateness of Lie affixed word (MinF1(2,112) = 36.7, p < .001 for the three conditions). They are slower to respond to a target word when it follows either a contextually inappropriate affixed word (289 ms) or a non-word (322 ms) than when it follows a morphologically complex word which is contextually appropriate (222 ms). RTs in these three conditions were significantly different from each other (at the .O1 level or beyond) on the Newman-Keuls statistic, using the MinF error 136 L.K. Tyler er al.

term. The same pattern of results was f~undfor both the derived and inflected words. This can be seen in Table 2, which shows the mean KTs in the three experimental conditions for derived and inflected words separately. BN's RTs were slower than those of the control subjects (BN: mean RT = 478 ms; controls = 278 ms). This slowing down of RTs is frequently observed after brain damage (Benton & Joynt, 1959). More importantly, his pattern of RTs were very differsnt from those of the control group, as can be .een in Table 2. An ANOVA showed that his monitoring latencies were not affected by either the contextual appropriateness of the suffixed word, (F(2,42) = 1.362, F = .26), or of the non-word. And this was the case for both derived and inflected words. None of the differences between the various experimental conditions were significant (on either the Newman-Keuls statistic or the t-test). We used the control group data to calculate the confidence limits for the differences between (a) the appropriate and inap~ropriateconditions, (b) the appropriate and non-word conditions and (c) the inapprooriate and non-word conditions for the derived and inflected sets separately. 11 the size of BN's difference scores is outside these limits, this is additional evidence that his performance differs from that of the control group. For both the derived and inflected forms, the size of the three sets of differences for BN was outside the confidence limits for the control group (p < .OS). These results clearly establish that BN is not sensitive to the contextual appropriateness of the inflected or derived test words, nor to the illegal combination of stem suffix in the non-words. Before exploring possible explana- tions for this insensitivity in greater detail, we first of all wanted to rule out the possibility that BN's poor performance was specific to the particular task we were using - the word-monitoring task. To evaluate this possibility we tested BN on the same materials, but this time we used a task which is standardly used in aphasia research - the gramnaticality judgement task - and which patients who, like BN, are classified as agrammatics have no prob-

Table 2. Mean RTs (ms)for control subjects and BN

Appropriate Inappropriate Non-word

Controls Derivations 218 286 319 Inflections 231 295 322 BN Derivations 545 489 526 Inflections 434 458 471 Processing distinctions between stems and offires 137

lems with (Linebarger, Schwartz, & Saffran, 1983). This is a task in which the patient is asked to indicate, after hearing each sentence, whether it is an acceptable sentence in their language. We tested BN and a group of control subjects on the grammaticality task. The performance of the control subjects was almost perfect in all conditions (ranging from 92% to 98%). However, BN's performance was very poor. He only correctly rejected 10% of the contextually inappropriate sentences and 33% of the sentences containing non-words, and he was equally bad with derived and inflected words. The results of the grammaticality judgement task' show that BN's insensitivity to the contextual appropriateness of a suffixed word in the monitoring task was not merely due to some peculiarity with the task or to some special difficulty he had with it. The results of both tasks together suggest that BN has problems evaluating the implications of both inflectionsl and derivational suffixes with respect to the sentential representation. This is why he is insensitive to their contextual appropriateness and why he is not affected by suffixed non-words. The next question we asked was whether this problem is specific to morphological affixes or whether it is part of a general problem which BN has in relating lexical contents to sentential representations. That is, does BN have particular problems evaluating the semantic and syntactic implic.ations of the suffix of a morphologically complex word, or is this just part of a more extensive problem in evaluating the syntactic and semantic properties of all types of morpheme? To investigate this possibility, we camed out Experi- ment 2.

Experiment 2: Stem violations The purpose of this study was to determine whether BN's problems are confined to the processing of inflected and derived suffixes or whether they are part of a more widespread problem with all types of morpheme. Once again, we used the word-tr.~ci~oringtask and our materials consisted of sentences containing inflected verbs (for additional methodological details see Marslen- Wilson, Brown, & Tyler, 1988; Tyler, 1985). This time, the inflection was always contextually appropriate whereas the stem could be semantically and/

'BN's poor performance on the grammaticality task does not reflect a general inability to perform the task. He scores well above chance on other experiments using the task - for example, on the %mematerials as described in Experiment 2. Here his judgements were aceurate 81% of the time. 138 L. K. Tyler et al.

Table 3. Experiment 2: Example stimuli

(a) The crowd was very happy. John was playing the @tar and ... (b) The crowd was very happy. John was brrrying the yitu and ... (C) The crowd was very happy. John was drinking the guihr and ... (d) The crowd was very happy. John was sleeping the gnkand ... - or syntactically inappropriate for the context. An example set of stimuli is given in Table.3. We constructed 32 sentence-pairs like the example in the table. The first sentence provided a niinimal context for the interpretation of the second sentence. The second sentence always took the following form. A subject noun phrase (NP) was followed by a verb which was followed by an object NP (the word which the subject had to monitor for). These 32 normal sentence pairs, like (a) constituted the baseline condition against which the other conditions could be evaluated. The other three conditions (b-cl) were constructed by varying the relationship between the verb and the target noun. In (b) the pragmatic implications of the verb make the following noun prag- matically (but not linguistically) anomalous. In (c) the noun violates semantic selection restrictions on the prior verb (Chomsky, 1965) and Ea (d) it violates strict suhcategcrisatien restrictions on the verb. The mean frequency (Kucera & Francis, 1967) of the verbs used in the four conditions was 27 per million (no violation), 19 (pragmatic violation), 24 (selection restriction violation) and 26 (categorial violation). We constructed four versions of the materials. Each version contained one quarter of the targets in each of the four conditions. Items in the four conditions were pseudo-randomly distributed across a version with 44 filler items interspersed between the test items. Timing pulses were placed at the onset of each target word on the non-speech channel of the tape. We used the word-monitoring task and the target word was always the object noun (e.g., "guitar" in the example set). To the extent that listeners are sensitive to these various types of linguistic and non-linguistic constraints generated by the verb stem as they interpret an utterance, RTs to the target noun should increase over the baseline condition (a) when they are violated. This is indeed what we found for a group of unimpaired listeners whose mean RTs in the four conditions are shown in Table 4. Their monitoring latencies were significantly slowed down (compared to the undisrupted condition) by all three types of anomaly, MinF (3,155) = 12.41, p i.091.~

'The difference between each type of violation and the undisrupted condition was significant on the Newman-Keuls statistic, using the MinF' error term. The difference between the pragmatic and semantic violations of 26 ms was only significant on a subject's analysis. Proc~ssingdiktinctions between stems and afaes 139

Table 4. Mean RTs (m)for BN and control subjects

Pragmatic Semantic Subcategory Undismpted violation violation violation

-- - -p------Controls 259 303 329 357 BN 487 531 537 570

- p-- --- p

BN's RTs, .,Ithough slower overall than the c?ntrn!s, showed essentially the same pattern (see Table 4). His RTs for each type of anomaly were significantly slower than for the undisrupted condition, F(3,93) = p c .OS. Newman-Keuls post-hoc tests showed that RTs for each type of anomaly were slower than for the undisrupted baseline condition (p c .05). These results show, first, that BN's responses in the word-monitoring task are sensitive to various types of linguistic violations. This means that BN's insensitid- ity to the types of lingtlistic violations we used in the first monitoring study did not merely reflect his inability to perform the task in the same way as unimpaired listeners. Second, the results show that when the inflection is consistent with the context, but the stem is contextually inappropriate in one way or another, BN's RTs increase. This indicates that he is sensitive to the contextual a~~ro~riatenessof the semantic and svntactic information carried L. . by stems and suggests that the problems he exhikted in t>e first study must be due to the inflectional and derivational momhemes. If BN has selective difficulties comprehending bound morphemes, what aspect of the comprehension process might be the scurce of this difficulty? To evaluate the various possibilities, we need to consiciar the range of processes involved in recognising spoken words. Listeners recz~nisespoken words by mapping the speech input onto a representation in the mestal lexicon of the phonological form of the word. We will assume here that this process occurs in the way described by the "cohort model" of lexical processing (Marsh-Wilson, 1984, 1987; Marslen-Wilson & Welsh, 1978; Tyler & Wes- sels, 1983). According to this model, some initial portion of the sensory input is mapped onto all those lexical representations with which it is compatible. These representations then become activated. As this mapping process continues, only those form representations which continue to match the sensory input remain activated. The activation levels of non-matching representations gradually decay. This process continues until there is only a single lexical reprzsentation which matches the sensory input. It is at this point (the "separation point") that a word is identified. The access of form representations is the route through which the semantic and syntactic content of a word can 140 L,K. Tyler et al.

be accessed. In principle, a patient may have a selective problem with either aspect of the process. The question we now need to ask about BN is whether his problems with bound affixes are caused by his inability either to access the form of a mor- pholo~icallycomplex word or to access its semantic andlor syntactic content. In Experiments 3 and 4, we examine BN's ability to access the phonological form of a morphologically complex word. These studies test whether he is able to use the speech input to make contact with the form representation in his mental lexicon of a morphologically complex word. In Experiments 5 and 6 we examine his ability to access the semantic and syntactic properties of morphologically complex words.

Experiment 3: Sdiied words. Gating task To determine whether BN has any difficulty with the processes involved in mapping the sensory input onto representations of the phonological form of a morphologically complex word, we used the gating task (Grosjean, 1980, Tyler & Wessels, 1983; 198.5) which measures the amount of sensory input listeners need to hear in order to identify the form of a word. Extensive research with unimpaired listeners has shown that lexical processing is highly efficient in the sense that listeners recognise a word at .the point at which it diverges from all other words sharing the same initial sound sequence (Marslen-Wilson, 1978, 1987; Tyler & Wessels, 1983, 1985). The gating task allows us to measure this point. In this task, subjects are presented with successively larger fragments of a word and after each fragment they either say or write down the word they think they are hearing. We can thus determine the point in the speech stream at which BN recognises a word and compare this with the point at which it is recognised by unimpaired listenen. We used a set of 18 bisyllabic test words. Twelve were morphologically complex and 6 were simple words. All the morphologically complex words were suffixed; half were derived and half inflected. The median frequency of the simple, inflected and derived words respectively was 9.5,5, and 12 (Fran- cis & Kucera, 1982). These stimuli were divided into two test blocks of 9 words, each preceded by a single practise word. The inflected words contained the suffixes [-ed], [ing], and [-es]. Each of these endings occurred once per block. A range of derivational suffixes were used. The morphologically simple words were a mixture of verbs and familiar, concrete nouns. These words were pseudo-randomly ordered and then read by a female native speaker of English. They were digitised at a sampling rate of 20 kHz Processing distinctions between stems and afFies 141

and segmented into fragments from word onset, with each fragment increas- ing in duration by 50 ms. To avoid splicing artefacts at the end of each gated segment, we smoothed the end by imposing a Hamming window on the final 5 ms of each fragment. The first fragment of each stimulus consisted of the first 50 ms of the word, the second consisted of the first 100 ms and so on throughout the entire word. The total number of fragments for each word ranged from 9 to 15 (mean = 12 fragments). The 6 control subjects (mean age 38 years) were tested in pairs in a quiet room. They were asked to write down after each fragment the word they thought they were hearing. BN was tested individually and the experimenter wrote down his responses. To assess how effectively listeners are able to map the sensory input onto representations of phonological form, we measure the point at which each subject correctly identifies the word without subsequently changing his or her mind. This is called the "isolation point" (IP; Grosjean, 1980, Tyler & Wes- sels, 1983,1985;Marslen-Wilson, 1984). In a number of studies, it has been shown that the IP correlates quite highly with the "separation point" as de- fined by the cohort model (Msrslen-Wilson, 1987; Tyler, 1984). In studies with unimpaired listeners, we also estimate the "recognition point", which is the point at which subjects correctly identify the word with some criteria1 degree of confidence. Since it is often not possible to obtain confidence rat- ings from aphasic patients, we only calculated the isolation points in this study. For each morphologically complex word, we measured the isolation point for the full form and for the stem. BN identified the total set of words at approximately the same point as the control subjects. An ANOVA calculated on the IPs for BN and the control group showed no significant differences, F(1,12) = 1.118, p = .0311. His mean IP for the full form of the simple and complex words combined was 322 ms, whereas the mean IP of the control group for the same words was 340 ms. Since the mean duration of the set of words used in the study was 600 ms, this means that both the control subjects and BN were correctly identifying a word after they had heard about half of it. A similar pattern was found for the derived and inflected words. BN's mean IP for the derived words was 380 ms, compared with 360 ms for the control group, and his PP for the inflected words was 350 ms, which was not significantly different from the control group's 320 ms. Similarly, he recognised the stem of both derived and inflected words at the same point as the control subjects (230 ms for controls vs. 255 ms for BN). There was one difference between BN and the control group. BN failed to identify two inflected words (fixes, sorted) by the end of the word. He did, however, correctly identify the stems of these words at the point where the control 142 L. K. Tyler et al.

subjects identified them, but failed to put the correct inflection on the stem. Instead, he used the inflection [kg]. These results from the gating task suggest that, for the most part, BN recognises both morphologicdiy simple and complex words in the same way as unimpaired listeners. That is, he maps the sensory input onto representations of lexical form and recognises a word at the same point at which it is recognised by the control group. Thus, he appears to be able to access the phonological form of a morphologically complex word in much the same way as unimpaired listeners.

Experiment 4: Suffixed WO& Auditory lexical decision task The results of the gating task told us about BN's general ability to access the form of morphologically complex words. In this fourth experiment, we examined his ability to correctly identify the same inflected and derived test words we had used in Experiment 1, and to discriminate these words from non-words. In that experiment, we found that BN was insensitive to the contextual appropriateness of a suffixed word and also to its legal morpholog- icai structure. The question we ask here is whether this is because he is unable to access the phonological form of those particular words. We do this by using the auditory lexical decision task. This enables us to determine whether BN can distinguish the suffixed real words from the morphologically complex non-words used in Experiment 1. The experimental stimuli consisted of the same derived and inflected lest words used in Experiment 1. There were 45 suffixed real words, which were a mixture of derived (e.g., wasteful) and inflected (e.g., causes) words, and 20 morphologically simple real wards (e.g., shadow). Each real word had a corresponding non-word. WC created morphologically complex non-words by substituting each word suffix with another suffix so that the resulting combination produced a non-word. For example, by changing the suffix on the real word ~vasreful,we created the non-word wastely. We turned morphologically simple words into non-words by changing the last syllable. So, for example, shadow became the non-word shadit. We included 140 filler items - half of which were words and half non-words. The real word fillers consisted of 25 conplex words (both prefixed and suffixed words) and 45 simple words. Thus, the number of simple and complex words were fairly evenly balanced with the combination of test adfiller items. The non-word fillers included simple words which had undergone phonetic changes ic varying positions in the word, and prefixed non-words. SO that subjects would only hear one member of a wordlnon-word pair at Processing distinctions between stems and afFies 143

a single testing session, we created two versions of the materials. Each version contained half the test materials - with an even number of real words and non-words - and all the filler items. Only one member of a wordlnon-word pair occurred in each version. Each version was recorded for auditory presen- tation to subjects. BN and a group of 6 control subjects (mean age 37 years) were presented with the list of stimuli over headphones. BN was asked to say whether each stimulus was a real word or a non-word, and his responses were recorded. The control subjects wrote down their responses on a score sheet. The control subjects were very accurate and made very few errors. They correctly accepted all the morphologically simple real words and rejected tne simple non-words. They made more errors on the suffixed words, failing to accept 4.76% of the derived real words and failing to reject 2% of the inflected non-words. Although BN made more errors than the controls, none of the differences were significant and his pattern of performance was essentially the same as the control grcup's. He was very accurate with simple words, only failing to accept one simple real word. Like the control subjects, he made errors with morpho1ogia:ly complex words. He failed to accept 6 (13%) of the morphologically complex real words and failed to reject 3 (7%) of the complex non-words. However, the difference in the number of errors between the two types of word was not significant (simple vs. complex real words: 2 = 1.45, p = .0735; non-words z = 1.418,~= .0778). BN also made more errors on inflected (17%) than derived words (9.5%), but again this difference was not significant (z = 13.7031, p = .242). Overall, BN was 90% correct in his ability to distinguish suffixed real words from non-words comprised of two real-word morphemes. On the whole, then, he was sensitive to the lexical status of suffixed words, suggesting that he does not have a serious problem accessing the form of a morphologically complex word. This suggests that his problems with suffixed words in Experiment 1 were not due to his inability to access the form of the morphologically complex words we used in the study. The results of Experiments 3 and 4 show that BN can efficiently map the sensory input onto the mental representation of the phonological form of a suffixed word, and that he can accurately discriminate real suffixed words from non-words. Taken altogether, then, the resi:!tq suggest that he does not have any difficulty in accurately identifying the phonological form of a suffixed word. If BN's insensitivity to the contextual appropriateness of morphological affixes is not due to problems in mapping the sensory input onto form-based lexical representations, it must be due to his inability to access the semantic and syntactic properties of bound morphemes. To test this hypothesis, we 144 L.K. Tyler et al.

designed two studies with inflected words in which we tried to separate those aspects of inflections which affect the grammatical structure of utterances (Experiment 5) and those which affect the semantic interpretation of an utterance (Experiment 6).

Experiment 5: Inflected words and their syntactic implications This study examined the role of inflected words as they relate to the syntactic structure of an utterance. It enables us to determine whether BN has problems in processing inflections when they serve a purely syntactic role in an utterance - in this case, by constraining the structural organization of an utterance. Test stimuli consisted of sentences with the structure: NP + AUX + [V+istlection] + NP. An example set of stimuli is given in Table 5. The second noun was the target word which the subject had to monitor for (we used the word-monitoring task as in Experiments 1 and 2). The nature of the inflection on the verb determined whether the verb was transitive or intransitive and this, in turn, set up a structural preference for either an NP or PP (prepositional phrase) to follow the verb. When the suffix is [-in& the verb is transitive and an NP is the preferred continuation. When the suffix is [-ed] the sentence is in the passive voice, the verb is intransitive and a PP is the preferred continuation. All past tense inflections were regular and syllabic, and both inflected forms were appropriate for the prior context. It is the form-class of the target word following the inflected verb which creates a syntactically legal or illegal continuation. A series of pre-tests established the syntactic appropriateness or inappro- priateness of the target noun.3 The question is whether BN can access the syntactic properties of the inflected verb and integrate them into the sentence representation. If he can, then RTs to targets following transitive verbs should be faster than those following intransitive verbs. The test sentence was either heard in isolation or it was preceded by a context sentence. We introduced this manipulation in order to see whether

-bne pre-test "etermined whether past tense verbs used in the inconsistent condition would be given a valid interpretation if the verb was parsed as an adjective (e.g., "They were baked potatoes ..."). The possibility of subjects interpreting the verb as an adjective had to be excluded otherwise items in the inconsistent condition would not necessarily be interpreted by the listener as being incorrect. A second pre-test established the strength of biases in the No-context condition. The number and animacy of the first Nt of the test sentence. together with the semantic features of the verb itself, may combine to provide a bias towards either a progressive or passive reading. or they may :e neutral and provide no specific bias. Items were categorised into one of the 3 bias categories according to subjects' responses. This did not interan with the effect of the consistency of the target (MinF' c l). Processing distinctions berween stems and affires 145

Table 5. Experiment 5: Example stimuli

- -. ~ . ~ - ~ ~ ~~ Context: Eric spent horn sitting by his easel.

(a) Appropridte target: He was painting boat6 down by the riverside. (b) Inappropriate target: He was pointed boab down by the riverside. -. .- ----p --

~able6. Mean RT in (m~)for the control .subjects and BN

Context NO-context - -. .. Correct Incorrect Correct Incorrect

Controls 228 282 255 322 BN 374 378 382 387

-. - p----~

the presence or absence of a prior context affects BN's ability to use the syntactic properties of the suffix. For the 36 test items we used in the study, the median frequency of the inflected verbs in the consistent and inconsistent conditions was 3 and 8.5, respectively. The median frequency of the target nouns was 27. We produced four versions of the 36 test items, such that each test item appeared in each version in only one of the four conditions. In each version, the test items were interspersed with 56 filler items, designed to counteract their regularities. The sequence of test and filler items was constant across the four conditions. They were prececr~dby 10 practice items. The items were recorded by 2 female native speaker of English onto one channel of a stereo tape. Timing pulses were placed on the non-speech channel of the tape, synchronous with the onset of the target word. The 24 young controls (ranging in age from 18 to 35 years) each heard gnly one version of the materials. BN was tested on all four versions, with one month separating each testing session. For the control subjects, whose mean RTs are displayed in Table 6, we found that RTs were faster when target words followed a transitive verb (247 ms) than an intransitive (282 ms) verb, MinF(1,41) = 25, p < .001. This suggests that the inflectional suffix functions to immediately constrain the syntactic structure of the utterance and this determines the syntactic appro- 146 L.K. Tylerefal.

Figure 1. Mean differences between consistent and inconsistent RTs for the contat and no contexf conditions in Experiment 5. Context El ~ocontext

priateness of the following noun. When it is inappropriate, RTs are longer than when it is appropriate. Figure 1, which presents the differences between the consistent and inconsistent conditions for the context and no-context conditions separately, shows that this effect was not modulated by the presence or absence of a discourse context (F < 1). RTs were faster to targets following transitive verbs whether or not there was a prior context sentence. The only effect which the discourse context had was to generally make it easier to identify the target word, MinF(1,27) = 4.57, p < .OS. BN's results are different from those of the control group. Most importantly, his latencies to respond to the target word are unaffected by the structural constraints generated by the transitivity of the verb (see Figure 1). When the verb is transitive and the target word is structurally appropriate his mean RT is 378 ms compared with 382 ms when the target noun is structurally inappropriate. An ANOVA comparing RTs in the appropriate and inappropriate conditions (collapsed across context) showed that these RTs were not significantly different from each other, F(1,19) = 0.480, p > .OS. Moreover, the difference between these two conditions was outside the confidence limits for tb control group (p > .OS). BN also showed no effect of the prior context, F(1,19) = 0.793, p > .OS. RTs were not significantly different whether the sentence was heard in isolation (385 ms) or preceded by a context sentence (376 ms) and there was no interaction between context and the appropriateness of the suffix. Processing distinctions between stems and aflxes 147

These results show that BN is insensitive to the structural implications of inflectional suffixes - at least of the type we used in this study. The next question we asked was whether BN is also insensitive to inflectional suffixes when they play a semantic, rather than syntactic, role in an utterance.

Experiment 6: Inflected words and their anapboric implications In English, the inflectional morphology does not only have a grammatical function. Because it marks tense and number. it also serves an anaphoric or deictic role. That is, inflectional morphemes can function to maintain time and reference within a dimme. For example, the present tense morpheme can either maintain the tense of the discourse or can signal a change of time. In this way, inflectional morphemes serve a semanticJpragmatic function in sentences. By examining BN's ability to process tense markers, we can determine whether he can evaluate the semantic implications of the inflectional morphology. The test materials (see Table 7 for an example) consisted of a context sentence followed by a continuation sentence. The context sentence was coil- structed so that it set up the present tense. The continuation sentence contained a verb which was either in the present or past tense. When the verb was in the present tense, verb tense and context tense were consistent. But when the verb was in the past tense, then the verb tense was inconsistent with the tense set up by the context sentence. A noun, functioning as a target word, immediately followed the tensed verb. If a listener is sensitive to inflectional morphemes marking tense, we expected monitoring latencies to be longer when there is a mismatch between context and verb tense. Two pre-tests ensured (a) that the test verb in its present tense form was appropriate for the tense established by the prior context whereas the past tense form was not, and (b) that the inappropriately inflected verbs (the past tense forms) were considered by subjects to be inappropriate. There were 20 test items. The median frequency of the targets in these

Table 7. Experiment 6: Example stimuli

)(a) Consistent tense: Alison isn't frightened of many things. These days she only hares bees and wasps if they crawl into her food. I(b) Inconsistent tense: Alison ishit frigintened of many things. These days she only hared bees and wasp if they crawl into her food.

-- ~ ~ p----p-- ~-~ ~ ~~ ~ ~~ 148 L.K. Tyleretal.

items was 18. Each test item appeared in two conditions: (a) where the context and verb tense were consistent; and (b) where they were inconsistent. Two versions of the materials were constructed so that each subject would hear each test item in only one condition. In each version the test items were pseudo-randomly interspersed with 60 filler sentences. The test items were counterbalanced for the syllabicity of the inflections, in case BN had special problems perceiving non-syllabic inflections. Half the test verbs had syllabic present tense forms (consistent items) and their corresponding past tense forms (inconsistent items) were non-syllabic. For the other 10 verbs, the present tense forms were non-syllabic and the past tense forms were syllabic. Monitoring RTs for the control subjects (see Table 8) were significantly slower when the verb tense was inconsistent with the context tense, but only when the inflection was syllabic (213 vs. 265 ms). When the inflection was not a full syllable, then RTs were not affected by the consistency of verb and context tense (223 vs. 234 ms ). On an ANOVA, the syllabicity by consistency interaction was significant, F(1,39) = 12.87, p < .001. The differences between the consistent and inconsistent conditions are shown in Figure 2. The lack of a consistency effect for the non-syllabic inflections may be due to the specificity of the bottom-up input. A non-syllabic inflecti~nis a more impoverished acoustic-phonetic stimulus than an inflection which is a full syllable, and the context may compensate for this minimal stimulus. There is amole evidence that listeners modulate their use of various kinds of information as the situation demands it. For example, less acoustic-phonetic information needs to be ~rovidedbv a s~eakerfor the correct identification of a word when an approphately consbai;ng context is available (e.g., Grosjean, 1980; Marslen-Wilson & Tyler, 1980; Miller, Heise, & Lichten, 1951). Such data suggest that the normal language-processing system is organized to allow the cooperative integration of different types of processing inforniarion with respect to the primary goal of interpreting an utterance within its discourse context. The fact that we do not see a consistency effect for non-syllabic

Table 8. Mean RT in (ms) for BN and the control subjects

~ ~ ~.. Syllabic Non-syllabic

~ ~ ~~~ --.-.. Consistent Inconsistent Consistent Inconsistent

~.- -..-~~ ~~-- . . .- Controls 213 265 223 234 BN 358 282 375 370

. . , ~ ~ ~ -~ -~ ~- ~~p~-~ Processing dktinctions benveen stems and affixes 149

Figure 2. Mean differences between the consistent and inconsistent conditions for the syllabic and non-syllabic items in Experiment 6.

Syllabic Moneyllabic '001

-100 ' Controls BN inflections may be another example of this general process. Supporting this hypothesis is the fact that RTs for both the consistent (223 ms) and inconsistent (234 ms) non-syllabic inflections are more similar to RTs for the consistent syllabic inflections (213 ms) than to RTs for the inconsistent sy1lab:c inflections (265 ms). This suggests that the tense of both the consistent and inconsistent non-syllabic inflections is being interpreted as consistent with the tense of the context. BN also shows no effect of the consistency of the non-syllabic inflections (375 vs. 370 ms) but, unlike the controls, his latencies are not slower when context and verb tense are inconsistent for the syllabic inflections (see Table 8 and Figure 2). In fact, his RTs are faster when the syllabic inflection is inconsistent with the context (282 ms) than when it is consistent (358 ms) with the context (t = 1.72, p = .OS). This difference between the two conditions was outside the confidence limits for the control group. For non-syllabic inflections, the difference beiween the two conditions was not significant (t = 0.13, p > .OS). These results suggest that BN is unable to integrate the anaphoric properties of the inflectional suffix (even when it is a full syllable) into the representation of the senteace context. 150 L.K. Tyler et al.

Conclusions The results of the six experiments described here show that BN has problems with the processing of inflectional and derivational suffixes and, moreover, they pinpoint the specific source of the problem. Experiments 3 and 4 established that BN's problems do not lie in his ability to access the phonological form of a morphologically complex word. In an auditory lexical decision task we found that he could accurately distinguish suffixed words from non-words, and in a gating task he recognised derived and inflected words after hearing the same amount of sensory input as the control subjects. The data from the gating study show that he is able to map the sensory input onto the appropriate representation of the phonological form of a morphologically complex word and that his processing of the speech input is, like unimpairec' listeners, highly efficient. Experiments 1, S and 6 showed that BN's ~roblemslie in his inability to integ;ate the semantic and syntactic of morphologica1:y complex words into the higher-level sentential representation. Whether a momholon- i -ally complex word generates inappropriate structural constraints (~x~ei- ~ii~ntS) or violates the tense of the discourse (Experiment 6) BN, unlike nnrmal controls, is insensitive to both types of violation. The results of Exper- iment 2 showed that this insensitivity to both the syntactic and semantic appropriateness of a morphologically complex word is confined to inflectional and derivational suffixes themselves. In this study, BN demonstrated that he had no problems in accessing the semantic and syntactic properties of stems and integrating them into the higher-level sentential representation. When the inflection was contextually appropriate but the relatibnship between the stem and the context was inappropriate, either on semantic or syntactic grounds, BN was as sensitive to these violations as unimpaired listeners. Hcwever, when the reverse situation held and the stem was contextually appropriate but the suffix was contextually inappropriate, BN was insensitive to the violations. Thus, it appears that BN has specific problems in integrating the syntactic and semantic aspects of the inflectional and derivational morphology into the contextual repre~entation.~ What conclusions follow from these results? First, there is 30 suggestion in these data that the underlying cause of BN's language comprehension disorder is due to a selective impairment in the representation of syntactic

?he way that BN processes morphologically complex words does not make clear predictions about bow he should process freestanding grammatical markers. If he treats them like stems (and certainly some types of free-standing grammatical morphemes could be considered to be more stem-like than affix-like), he should pcrform normally: if he treats them like aftixes, he should have problems. Processing disrinctions between srems and affixes 151

knowledge associated with bound morphemes. For this to have been the case, he should have shown selective difficulties with those bound morphemes which are thought to primarily play a syntactic role in utterances (inflectional morphemes), and with the syntactic, as opposed to the semantic, properties of inflectional morphemes. Instead, he showed similar degrees of impairment for both the derivational and inflectional morphology, and he showed no sensitivity to either the syntactic or the semantic properties of inflectional morphemes. In his across-the-board problem with both types of suffixed words, BN is unlike some other ;-~tientsreported in the literature, who show a selective deficit fcr the inflectional morphology (Miceii & Caramazza, 1987; Tyler & Cobb, 1987) and appear to comprehend derived words normally. This is particularly interesting given that his spontaneous speech is so similar to that of the patient reported in Tyler and Cobb (1987) who has difficulty comprehending only inflected forms. It adds support to the claim that patients typically labelled as "agrammatic" ' 1 their speech production may have different underlying comprehension deficts (Martin et al., 1989; Goodglass & Menn, I985), and calls into question theories which describe the disorder as a central syntactic deficit. BN appears to analyse both derived and inflected words in ihe same way -by recognising the stem and not processing the suffix. This is not to say that once he recognises a stem he does not extract any information from the rest of the word. Experiments 3 and 4 show that the acoustic-phonetic input corresponding to the entire full-form of a morphologically complex word is mapped onto its form-based lexical representation. So, the entire phonological form of the word is analysed at the lexical level. What seenls to be happening is that only the syntactic and semantic content of the stem is accessed and integrated with the context. Tile semantic and syntactic implications of the suffix are never evaluated. This implies that the content of stems and affixes can be evaluated against the context independently of each other. This, in turn, suggests that stems and suffixes must be separately represented at some level in the mental lexicon (cf. Tyler & Marslen-Wilson, 1986). Moreover, since BN shows the same pattern of results for both derived and inflected words, we conclude that this may hold for both types of morphologically complex words. How- ever, we can only make this c!aim with any certainty for the particular type of suffixed words we used in our experiments. These were words in which the relationship between the stem and the suffix was both phonoiogically and semantically transparent. They were phonologically transparent in that the suffix did not cause major phonological changes to the stem, and they were semantically transparent in that the meaning of the full-form was readily 152 L.K. Tyler er al.

recoverable from the meaning of the stem. This was true for both the derived (e.g., enjoyab~elenjoyment)and the inflected words (e.g., completedlcomplet- ing). BN's data suggest that suffixed words with these properties may be represented in the mental lexicon as stems plus their associated affixes - whether or not they are derivecl or inflected. If this is the case, then inflected and derived words may not 11tct:ssarily have different types of representation. The question which remains open is whether morphologically complex words which are semantically opaque (e.g.. department) are represented in the same way as those which sre semantically transparent -.that is, in terms of a stem and its associated affixes. (This only applies to derived words, since inflected words are always semantically transparent.) If they are not, then the way in which a complex word is represented in the mental lexicon may depend more upon whether its meaning is readily recoverable from its component parts than upon whether it is derived or inflected.

References

Anderson. S. (1982). Where's morphology? Linguinic Enquiry. 13.571-612. Benton. A.L.. & Joynt. R.J. (1959). Reaction time in unilateral ierebral disease. Confiiu fleurologica, 19. 247-256. Berndt. R.. & Caramaua. A. (1980). A redefinition of the syndrome of Bmca's aphasia: Implications for a neuropsychologiwl model of language. Applied Psycholinguistics. 1.225-278. Bishop. D. (1982). TROG: Test for the Reception of Grammar. Abingdon. U.K.: Thomas Leach (for the Medical Research Council). Butterworth. B. (1983). Lexical representation. In B. Butterworth (Ed.;. Language Production (Vol. 2). London: Academic Press. Chomsky. N. (1965). Aspects of the theory of syntax. Cambridge. MA: MIT Press. Cooper. W.E.. & Zurif. E. (1983). Aphasia: Information processing in language production and reception. In B. Butterworth (Ed.). Language production (Vol. 2). London: Academic F'ress. Je Renzi. E.. & Faglioni. F. (1978). Normative data and screening power of a shortened version of the Token Test. Cnrte.~.14. 41-49, Devillien. J.G. (1978). Fourteen grammatical morphemes in acquisition and aphasia. In A. Caramazza & E. Zurif (Eds.). Language acquisition and language breakdown: Parallels and divergences. Baltimore: Johns Hopkins Clnivenity Press. Eling, P. (1986). Recognition of derivatims in Broca's aphasia. Brain and Language, 28, 346-356. Francis, W.. & Kucera. H. (1982). Frequency Analysis of English Usage: Lexicon and Grammar. Boston: Houghton MiMin. Garrett. M. (1980). Levels of processing in sentence production. In B. Butterworth (Ed.), Languageproduc- tion (Vol. 1). London: Academic Press. Gwdenough. C.. Zurif. E.. & Weintraub. S. (1977). Aphasia' attention to grammatical morphemes. Lan- guage and Speech. 20. 11-19, Goodglass. H.. & Berko. 3. (1%0). Agrammatism and inflectional morphology in English. Journal of Speech and Hearing Research. 3,257-267. Goodglass. H.. Gleason, J.. Ackerman. N.. & Hyde, M. (1972). Some linguistic structures in the speech of a Broca's aphasic. Conex, 8, 191-212. Processing distinctions between stems and affixes 153

Goodglass, H., & Kaplan, E. (1972). The ASS~~Y?I~FZ?of Aphasia and Related Disorders. Philadelphia: Lea & Febiger. Goodglass, H., & Menn, L. (1985). Is agrammatism a unitary phenomenon? In M.L. Kean (Ed.), Agram- matism. New York: Academic Press. Grodzinsky, J. (1984). The syntactic characterisation of agrammatism. Cognition, E&99-120. Grosjean, F. (1980). Spoken word recognition processes and the gating paradigm. Perception and Psycnophysics, 28,267-283. Grossman. M., Carey, S., Zurif, E., & Diller, L. (1986). Proper and common nouns: %rr, class judgements in Broca’s aphasia. Brain and Language, 28, 114-125. Kucera, H., & Francis. W. (1967). Computational analysis of present-day American English. Providence, RI: Brown University Press. Linebarger, M., Schwartz, M., & Saffran, F. (1983). Grammatical judgements in so-called agrammatic aphasics. Cognition, 13,36!-392. Lukatela, K., Grain, S., & Shankweiler, D. (1988). Sensitivity to inflectional morpllology in agrammatism: Investigation of a highly inflected language. Brain and Language, 33; l-15. Marslen-Wilson, W.D. (1984). Function and process in spoken word recognition. In H. Bouma and H. Bouhuis (Eds.), A:tcwion and Performance X: Contrc! gf Language Processes. Hillsdale, NJ: Erlbaum. Marslen-Wilson, W.D. (1987). Fun&onal parallelism in spoken word recognition. Cbgnition, 25, 71-102. Marslen-Wilson, W.D., Brown, C., LyrTj!er, L.K. (1988). Lexical representations in spoken language comprehension. Language and Cognitive Processes, 3. l-17. Marslen-Wilson, W.D., & Tyler, L.K (1980). The temporal structure of spoken language understanding. Cognition, 8, l-71. Marslen-Wilson, W.D., & Welsh, A. (1978). Processing interactions and lexical access (during word recognition in continuous speech. Cognitive Psychology, 10, 29-63. Martin, R., Blossom-Stach, C., & Feher, E. (1989). Syntactic loss versus processing deficit: An assessment of two theories of agrammatism and syntactic comprehension deficits. Cognition, 32, 157-191. Miceli, G., & Caramazza, A. (1987). Dissociation of inflectional ar:d derivqtional morphology. Dissociation of inflectional and derivational morphology. Reports of the Cognitive Neuropsychology Laboratory, Johns Hopkins University. Miller, G., Heise, G,, & Lichten, W. (1951). The intelligibility of speech 2s a function of the context of the test materials. Journal of Experimental Psychology, 41, 329-336. Patterson, K. (1979). What is right with deep dyslexic patients? Brain and Larguage, 8, 111-129. Stemberger, J., & MacWhinney, B. (1986). Frequency and the lexical storage of regularly inflected forms. Memory and Cognition, 14, 17-26. Tyler, L.K. (1984). The structure of the initial cohort: Evidence from gating. Perception arzd Psychophysics, 36,417-427. Tyler, L.K. (1985). Real-time comprehension processes in agrammatism: A case study. Brain and Language. 26, 259-275. Tyler, L.K. & Cobb, H. (1987). Processing boi:nd grammatical morphemes in cantext: The case of an aphasic patient. Language and Cognitive Processes, 2, 245-262. Tyler, L.K., & Marslen-Wilson, W.D. (1986). The effects of context on the recognition of polymorphemic words. Journal of Memory and Language, 25, 741-752. Tyler, L.K., & Wessels, J. (1983). Quantivng contextual contributions to word recognition processes. percep tion and Psychophysics, 34,405-420. Tyler, L.K., & Wessels, J. (1985). Is gating an on-line task? Evidence from naming latency data. Pe=ption and Psychophysics, 38, 217-222.