agrammatism is a syntactic deficit following damage to the left hemisphere, usually in Broca's area and its vicinity (Zurif, 1995). The traditional view concerning speech production in agrammatism was that syntactic ability is completely lost, and agrammatic aphasics rely on nonlinguistic strategies to concatenate words into sentences (Goodglass, 1976; Berndt and Caramazza, 1980; Caplan, 1985), or that all functional elements are impaired in agrammatic speech production (Grodzinsky, 1990; Ouhalla, 1993). However, in recent years empirical evidence has accumulated to suggest that the deficit is actually finer-grained.

Speech production in agrammatism shows an intricate and intriguing pattern of deficit. Individuals with agrammatism correctly inflect verbs for agreement but substitute tense inflection; they produce well-formed yes/no questions (in some languages), but not Wh questions; they can produce untensed embedding but not full relative sentences, and coordination markers but not subordination markers. The tree pruning hypothesis (TPH) was suggested by Friedmann (1994, 2001) and Friedmann and Grodzinsky (1997) to account for these seemingly unrelated deficits and for the dissociations between spared and impaired abilities within and across languages. The TPH is a linguistic generalization, formulated within the generative grammar framework, and was suggested to account for production only. (For a syntactic account of agrammatic comprehension, see trace deletion hypothesis; Grodzinsky, 2000.) According to the TPH, individuals with agrammatic aphasia are unable to project the syntactic tree up to its highest node, and their syntactic tree is “pruned.” As a result, syntactic structures and elements that require the high nodes of the tree are impaired, but structures and elements that involve only low nodes are preserved (Fig. 1).

Figure 1..  

A syntactic tree (Pollock, 1989). The arch represents a possible impairment site according to the tree pruning hypothesis. Nodes below the arch are intact and nodes above it are impaired.

According to syntactic theories within the generative tradition (e.g., Chomsky, 1995), sentences can be represented as phrase markers or syntactic trees. In these syntactic trees, content and function words are represented in different nodes (head nodes and phrasal nodes) (Fig. 1). Functional nodes include, among others, inflectional nodes: an agreement phrase (Agr_sP), which represents agreement between the subject and the verb in person, gender, and number, and a tense phrase (TP), representing tense inflection of the verb. Finite verbs move from V⁰, their base-generated position within the VP, to Agr⁰ and then to T⁰ in order to check (or collect) their inflection. Thus, the ability to correctly inflect verbs for agreement and tense crucially depends on the AgrP and TP nodes.

The highest phrasal node in the tree is the complementizer phrase (CP), which hosts complementizers such as “that” and Wh morphemes such as “who” and “what” that moved from the base-generated position within the VP. Thus, the construction of embedded sentences and Wh questions depends on the CP node being intact and accessible.

Crucially, the nodes are hierarchically ordered in the syntactic tree: the lowest node is the verb phrase, the nodes above it are the agreement phrase and the tense phrase (in this order according to Pollock, 1989), and the complementizer phrase is placed at the highest point of the syntactic tree.

The TPH uses this hierarchical order and suggests that in agrammatism, the syntactic tree is pruned from a certain node and up. Persons with agrammatic aphasia who cannot project the syntactic tree up to the TP node are not able to produce structures that require TP or the node above it, CP. Persons with agrammatic aphasia whose tree is pruned at a higher point, CP, are unable to produce structures that involve the CP. Importantly, nodes below the pruning site are intact, and therefore structures that require only low nodes, such as AgrP and VP, are well-formed in agrammatic production.

How does tree pruning account for the intricate pattern of loss and sparing in agrammatic production? The implications of tree pruning for three syntactic domains—verb inflection, embedding, and question production—are examined here.

Inflections are impaired in agrammatic production, but in a selective way. In Hebrew and in Palestinian Arabic, tense inflection was found to be severely impaired, whereas agreement inflection was almost intact in a set of constrained tasks such as sentence completion, elicitation, and repetition (Friedmann, 1994, 2001; Friedmann and Grodzinsky, 1997). Subsequent studies reported a similar dissociation between tense and agreement in Spanish, Dutch, German, and English. If tense and agreement reside in different nodes in the syntactic tree, and if TP is higher than AgrP, this dissociation is explained by tree pruning: the TP is inaccessible, and therefore tense inflection is impaired, whereas the AgrP node, which is below the pruning site, remains intact, and subsequently agreement inflection is intact. This can also account for the findings from German and Dutch, according to which individuals with agrammatic aphasia frequently use nonfinite verbs in sentence-final position, instead of the required fully inflected main verbs in second position (Kolk and Heeschen, 1992; Bastiaanse and van Zonneveld, 1998). The lack of TP and CP prevents persons with agrammatic aphasia from moving the verbs to T to collect the required inflection, and to C to second position, and therefore they produce the verbs in a nonfinite form, which does not require verb movement to high nodes. Verbs that do not move to high nodes stay in their base-generated position within VP, which is the sentence-final position (see Friedmann, 2000).

Individuals with agrammatic aphasia are also known to use only simple sentences and to avoid embedded sentences. When they do try to produce an embedded sentence, either a relative clause (such as “I saw the girl that the grandmother drew”) or a sentential complement of a verb (“The girl said that the grandmother drew her”), they fail, and stop before the complementizer (the embedding marker “that,” for example), omit the complementizer, use direct instead of indirect speech, or produce an ungrammatical sentence (Menn and Obler, 1990; Hagiwara, 1995; Friedmann, 1998, 2001). The difficulty posed by embedded constructions is explained by tree pruning, as full relative clauses and sentential embeddings require the CP, and when the CP is unavailable, these structures are impaired. Interestingly, embedded sentences that do not require the CP, such as reduced relatives (“I saw the boy crying”), are produced correctly by individuals with agrammatism.

Tree pruning and the inaccessibility of CP cause a deficit in another important set of structures, questions. Seminal treatment studies by Shapiro, Thompson, and their group (e.g., Thompson and Shapiro, 1995) show that persons with agrammatic aphasia cannot produce well-formed Wh questions. Other studies show that in English, both Wh and yes/no questions are impaired, but in languages such as Hebrew and Arabic, Wh questions are impaired but yes/no questions are intact (Friedmann, 2002). Again, these dissociations between and within languages are a result of the unavailability of the CP: agrammatic aphasics encounter severe difficulties when trying to produce Wh questions because Wh morphemes (who, what, where, etc.) reside at CP. Yes/no questions in English also require an element at CP, the auxiliary (“Do you like cream cheese”?), and are therefore impaired too. Yes/no questions in Hebrew and Arabic, on the other hand, do not require any overt element in CP (“You like hummus?”), and this is why they are produced correctly.

The tree pruning hypothesis is also instrumental in describing different degrees of agrammatism severity. Clinical work shows that some individuals with agrammatism use a wider range of syntactic structures than others and retain more abilities, such as verb inflection, whereas other individuals use mainly simple sentences and substitute inflections. It is possible to characterize the milder impairment as pruning at a higher site in the tree, at CP, whereas the more severe impairment results from pruning at a lower position, TP. Thus, the more mildly impaired individuals show impairment in Wh questions and embedding, but their ability to inflect verbs for both tense and agreement is relatively intact. More severely impaired individuals, who are impaired also in TP, show impaired tense inflection in addition to impairment in questions and embedding. In both degrees of severity, agreement inflection is intact. Crucially, no individual was found to exhibit a deficit in a low node without a deficit in higher nodes. In other words, there was no TP deficit without a deficit in CP, and no deficit in AgrP without a deficit in TP and CP.

Finally, tree pruning provides a principled explanation for the effect that treatment in one syntactic domain has on other domains. Thompson and Shapiro, for example, report that following question production treatment, their patients started using sentential embedding. This can be explained if treatment has enhanced the accessibility of the syntactic node that is common for the two structures, CP. Similarly, the decrease in verb omission that has been reported to accompany tense inflection improvement can be explained by enhanced accessibility to the inflectional node TP.

Current linguistic theory thus provides a useful toolbox to account for the complicated weave of spared and impaired abilities in speech production in agrammatic aphasia. The hierarchical structure of the syntactic tree enables an account of the highly selective syntactic deficit in agrammatic production in terms of syntactic tree pruning.