Some aphasic syndromes implicate grammar. Known for almost a century to be grammatically impaired in speech production (see syntactic tree pruning), individuals with Broca's and Wernicke's aphasia are now known to have receptive grammatical deficits as well. The trace deletion hypothesis (TDH) is a collection of ideas about the proper approach to linguistic deficits subsequent to brain damage in adults, particularly Broca's aphasia. Support for a grammatical interpretation of aphasia comes from a dense body of research from many laboratories and varied test paradigms, administered to large groups of aphasic speakers of a variety of languages.

This article explains why a precise characterization of receptive deficits in aphasia is important, and how we came about to know it. It first aims to demonstrate that Broca's and Wernicke's aphasia (and perhaps other syndromes) are syntactically selective disorders in which the line dividing impaired from preserved syntax is fine and amenable to a precise characterization.

The comprehension of sentences containing grammatical transformations is impaired in aphasia. Transformations, or syntactic movement rules, are important intrasentential relations. In their various theoretical guises, these rules are designed to explain dependency relations between positions in a sentence. As a first pass, we sacrifice precision for clarity, and say that transformationally moved constituents are found in noncanonical positions (e.g., a subject in a passive sentence, The teacher was watched___by the student; a questioned element in object questions, Which man did Susan see ___?). Each of these sentences has a nontransformational counterpart (e.g., The student watched the teacher; Which man saw Susan?). Roughly, in transformationally derived sentences, a constituent is phonetically present in one (bolded) position but is interpreted semantically in another, “empty” position (___), which is annotated by a t (for trace of movement) and serves as a link between the semantic role of the moved constituent and its phonetic realization. Traces are therefore crucial for the interpretation of transformational sentences.

When aphasic comprehension is tested on the contrast between transformational and nontransformational sentences, a big difference is found: Broca's aphasics understand active sentences, subject questions, and the like normally, yet fail on their transformational counterparts. This has led to the claim that in receptive language, Broca's aphasics cannot represent traces of movement. These traces are deleted from the syntactic representations they build, hence TDH. This generalization helps localize grammatical operations in the brain. Furthermore, the highly selective character of this deficit has major theoretical ramifications for linguistic theory and the theory of sentence processing.

How exactly does the TDH work? Let us look at receptive tasks commonly used in research on brain-language relations. These are sentence-picture-matching (SPM), grammaticality judgment (GJ), and measurements of reaction time (RT) during comprehension. We can consider the SPM method first. A sentence containing two arguments is presented (e.g., The student watched the teacher); it is “semantically reversible”—the lexical content does not reveal who did what to whom. The task, however, requires exactly that: subjects are usually requested to choose between two action pictures in which the roles are reversed (student watching teacher, teacher watching student). Reliance on syntax is critical. On being given a series of such sentences, Broca's aphasics perform at above chance levels on nontransformational sentences and at chance levels on transformationally derived sentences.

What do such findings mean? Standard principles of error analysis dictate that a binary choice design allows for three logically possible outcomes: above-, below-, and at-chance performance levels. Above-chance performance is virtually normal; below-chance amounts to systematic reversal of roles in SPM; at-chance means guessing, as if the subject were tossing a coin prior to responding. Indeed, the response pattern of Broca's aphasics on transformational sentences resembles guessing behavior. That said, the burden on the TDH is to provide a deductive explanation of why left anterior lesions, which allow normal comprehension of sentences without transformations, lead to guessing when transformational movement is at issue.

The foregoing was a simplified introductory discussion. In reality, things are somewhat more complicated. The deletion of a trace in certain cases does not hinder performance in aphasia. Our characterization needs refinement. The leading idea is to view aphasic sentence interpretation as a composite process—an interaction between incomplete syntax (i.e., representations lacking traces) and a compensatory cognitive strategy. The interpretation of moved constituents, as we saw, depends crucially on the trace; without traces, the semantic role of a moved constituent cannot be determined. Moved constituents (bolded) are thus uninterpretable. A nonlinguistic, linear order–based cognitive strategy is invoked to try and salvage uninterpreted NPs %%NP₁ = agent; NP₂ = theme%%. In English, constituents moved to a clause-initial position will thus be agents. In certain cases (e.g., subject questions), the strategy compensates correctly: in the subject relative the man who t loves Mary is tall, the head of the relative clause, the man, is moved, and receives its semantic interpretation (or thematic role) via the trace. A deleted trace blocks this process, and the strategy is invoked, assigning agenthood to the man and yielding the correct semantics: NP₁ (the man) = agent by strategy, and NP₂ (Mary) = theme by the remaining grammar. In other cases the TDH system results in error: In object relatives—the man who Mary loves t is tall—the agent role assigned by the strategy (acting subsequent to trace deletion) gives rise to a misleading representation: NP₁ (the man) = agent by strategy, and NP₂ (Mary) = agent by the grammar. The result is a semantic representation with two potential agents for the depicted action, which leads the patients to guessing behavior. The selective nature of the aphasic comprehension deficit is captured, which is precisely what the TDH is designed to explain.

Languages with structural properties different from English lend further support to the TDH. Thus, Chinese, Japanese, German and Dutch, Spanish, and Hebrew have different properties, and the performance of Broca's aphasics is determined by the TDH as it interacts with the particular grammar of each language. Japanese active sentences, for example, have two configurations: [Taro-ga Hanako-o nagutta (Taro hit Hanako)—Subject Object Verb], [Hanako-o Taro-ga t nagutta—Object Subject t Verb]. These simple structures mean the same and are identical on every dimension except movement. Broca's aphasics are above chance in comprehending the former and at chance level on the former, in keeping with the TDH. A similar finding is obtained in Hebrew. In Chinese, an otherwise SVO language like English, heads of relative clauses (annotated by the subscript h) follow the relative (unlike English, in which they precede it, as the example shows):

This structural contrast leads to remarkable contrastive performance in Broca's aphasia: in Chinese subject relatives (1), the head of the relative (the cat) moves to the front of the sentence, lacks a role, and is assigned agent by the strategy, which leads to a correct representation in which the cat indeed chases the dog. In Chinese, the head (mau) also moves, yet to sentence-final position, and the strategy (incorrectly) assigns it the theme role. This representation now has two themes, the dog and the cat, and the result is guessing. Similar considerations that hold for the object relatives are left to the reader. English and Chinese thus yield mirror-image results, which correlate with a relevant syntactic contrast between the two languages. Further intriguing cross-linguistic contrasts exist.

Moving to other experimental tasks, we encounter remarkable cross-methodological consistency. When the detection of a violation of grammaticality critically depends on traces, the TDH predicts disability on the part of Broca's aphasics. This is indeed the case. When given GJ tasks, they show differential performance between sentences with and without traces, again in keeping with the TDH. Finally, the rich literature on timed language reception in Broca's aphasia (RT) suggests that on-line computation of trace-antecedent relations is compromised.

This type of deficit is not restricted to Broca's aphasia. On at least some tests, Wernicke's aphasics perform like Broca's aphasics. There are contrasts between the two groups, to be sure, yet contrary to past views, there are overlapping deficits. Although this new picture is just beginning to unfold, independent evidence from functional imaging of normal populations (fMRI) supports it. Both Broca's and Wernicke's regions of the healthy brain are involved in transformational analysis, although likely in different ways.

This series of cross-methodological findings from different languages and populations, and the TDH as a generalization over them, have a host of important theoretical implications. They show that the natural classes of structures that linguists assume have a firm neurobiological basis; they afford an unusual view on the inner workings of the human sentence processing device; they connect localized neural tissue and linguistic concepts in a more detailed way than ever before. Finally, there appears to be clinical (therapeutic) value to the view that Broca's aphasia and Wernicke's aphasia are grammatically selective: although currently experimental, preliminary results suggest that therapeutic methods guided by this view are somewhat efficacious.