Passive and Raising

Passive Voice and Raising

Linguistics 322

Intermediate Syntax

Voice is a grammatical category, an operator, that has no directly link to conceptual structure. Voice is a complex issue, but it appears to mark a prominent object that the speaker chooses to make prominent. In English there are two voices: active and passive. The passive voice is the marked voice: [+Passive], and the active voice is the unmarked or default voice: [-Passive]. Virtually every verb may occur in the active voice--indeed some of them must:

1.    John ate the potato.

2.    Some birds are flying.

3.    Henry resembles Hank's father.

The passive voice is restricted to transitive verbs. There is a small class of transitive verbs that cannot be marked in the passive voice.

4.     The potato was eaten by John.

5.     *Be flying some birds.

6.     *Hank's father is resembled by Henry.

In the grammatical propositional structure, voice like tense and moods is an operator. The first feature set is [±Pass] (Passive). [-Pass] is the default represent the active voice The passive voice is represented as:

7. [VOICE [±Pass]] < VP>.

As far as it is known voice always takes a verb phrase as its argument.

The logical structure for (4) is the following, ignoring the relevance and aspect operators which turn out to be null here:

8.

Let's start with the voice operator. The lexical entry includes information that [+Pass] needs a verbal host. A copy of the voice operator is made and lowered and adjoined to the verb V, which VOICE governs. V is not spelled out until after features have been copied downwards:

(9).

T is marked with "V\____" indicating that it needs a host. One of the restrictions of English morphology is that a host can host only one inflectional form. Some languages can host two or more. We will return to the host problem at the final stage called spell-out.

The function of the passive voice is to bring the object into prominence instead of the subject. If the object NP is marked for the accusative Case, then it cannot raise and adjoin to TP (or raise to Spec-T), since T assigns the nominative Case to the left. There is a general principle that prohibits assigning more than one Case to a NP. It is uncertain at this time whether this is a universal principle of grammar or whether this is a language specific parameter:

(10) Principle: *NP, if NP assigned more than one Case.

In order for the object to raise to T, the verb cannot assign the accusative Case to it. Chomsky uses the term lift Case when a verb fails to assign Case to its complement. We can attribute Case-lifting here the passive feature.

(11) Rule: When a Verb contains the feature [+Pass], it fails to assign the accusative Case to its complement.

On the one hand we have to raise one of the NP complements to the subject position, and other the other hand both NPs need Case. If the agent, or secondary complement, were raised, there would be no difference between the active and the passive, except that there would be one NP still unmarked for Case:

(12) *John was eaten the potato.

If the object is raised, it becomes prominent:

(13) *The potato was eaten John.

Of course, this meets with the objective to make John prominent.

Just how does this come about? This is accomplished, we theorize, through targeting (see Voice and Raising ). The target builds a shield around the targeted NP, which blocks Case agreement with the verb head:

(14)

A link is established between the subject position S, marked with the feature [+PT] and the NP marred with the same feature; al the features dominated by the NP are copied upwards to the subject position:

(15)

What about the agent NP which is unmarked for Case? English and other natural languages have a card up their sleeve: It inserts a dummy preposition and adjoins it to the left of the Caseless NP as a last resort. The dummy preposition assigns the accusative Case to its complement. There are at least three dummy Ps in English: by, with, of. The dummy preposition here is the preposition by:

(15) The potato was eaten by John.

Dummy prepositions cannot be modified:

(16) *The potato was eaten right by John.

Note that there is another reading of (15) where by can be modified by a degree word. In this reading, by means 'next to.'

(17) The potato was eaten right next-to John.

This means that (15) is ambiguous: John is either the agent or the object of the locative preposition by.

Diagram (17) shows the result of raising to subject and the insertion of the morpheme (BY). If it is true that all phonetic insertion occurs last, then {BY} isn't spelled out here. Note that {BY} has no semantic content and hence no propositional form *BY, since {BY} is a dummy morpheme.

T remains bound. It still needs a host. Since it cannot lower, how can the structure be saved? We are already familiar with the strategy of dummy insertion. It applies as a last resort. In T immediately c-commands Pass (or T is a sister to PassP), the dummy auxiliary verb {be} is inserted merging with T.

(18)

In the penultimate stage of the logical derivation, subject-verb agreement applies. Subject-verb agreement cannot apply until the subject NP has been determined. Originally, the head of SP, the subject, governs VP and V. When the features of NP are copied to the subject position, the link holds between the head NP of the subject and the head that it governs.

What triggers subject-verb agreement? This is appears to be the same type of agreement we found within NPs. Let us assume this to be the case. Suppose we mark the features of T with the blank features of person and number:

(19)

Value Feature

T (ease) Category

[±Past] Tense

Person

Personal

S (subject) is the head of SP and is a non-argument. It must be marked in some way. So far, the features of the prominently marked NP are copied to S. S is now a projection. A governing link is established between the head of the project and the head of the governed category:

(20)

The feature of plural is copied to N from Qu, and the features of Plural and Person are copied from N to T--subject-verb agreement.

And in the final stage, spell-out applies producing phonological and orthographic form of each category (we show the orthographic form only). The lexical entry for eat contains the information that the non-progressive participle is spelled out as the suffix '-en' and with the base form of the verb: eat+en. Thus V must split into V[stem]+Pass[suf]. The following figure shows both percolation and phonological insertion:

(21)

The passive in English contains at least one interesting detail. The agent argument is a secondary complement. Secondary complements are always optional, unlike primary complements. This means that the agent may be omitted:

(20) The potato was eaten.

The problem is how to represent this in the grammatical propositional form. Perhaps this is a good time to discuss this a little further. The grammatical proposition may be what Chomskyites call Logical Form. It represents the logical structure of the grammar. This is distinct from the semantic component. The semantic component is is where true semantics lies. This is where the truth and falseness of a proposition lies. The speaker knows whether a proposition is true or false. He projects (maps) a message into the grammar (see grammar). The grammar determines which grammatical proposition the grammar must contain. The message is filtered into this system. We have been starting with Logical From (the grammatical propositions). LF determines the grammatical categories that exist in the language, such as mood, tense, and voice in the verb and number in the noun.

There are many transitive verbs that require both a subject (agent or experiencer) and a direct object (theme or patient). In the active voice, the agent or experiencer may be omitted in the grammar. But they are implied. The implication is in the \semantics. In (18), for example, the theme is required, but the agent is optional. In semantics there is always an agent, even if it is unknown what the agent is. The agent is sometimes called an external argument. This we call a secondary complement. Thus the grammar of English which arguments can or must be stated in which voice.

The situation becomes more complex in examples of the following sort:

(21)     Mary was run over by a car.

(22)     Mary was run over with a car.

(23)     A car ran over Mary.

(24)     Someone ran over Mary with a car.

(21) implies that there is no perceivable or known agent. The car could have slipped its brakes. There is no perceivable agent. (22) implies that there is an agent, but the agent is not expressed. Either the speaker does not know who the agent is or he does not wish to divulge the identity of the driver of the car. In the passive (22) the agent argument may be explicit or implicit. If the external argument in the active voice is not an agent (23), then the agent is simply not denoted. Someone in (24) gives us the same amount of pragmatic information as the lack of an overt agent in (22)--someone does imply a human agent, which is not explicit in (22). Pragmatically, only a human can drive a car and run over someone. If a dog knocks the gear selector into neutral and the car starts to roll and runs over Mary, do we say that the dog is the agent? Perhaps so. If so, then (22) and (24) are not exactly synonymous but an agent is implied in both cases.

How do we mark implicit arguments? Let us mark it 'X' in the following way:

(25) <agent: X>

Note we could write someone as

(26) <agent: [+Human]>.

(26) seems to match information content of someone when it is an agent. Thus we can mark voice and the propositional content of (21) through (24) as:

(27)      [+Pass] RUN OVER <patient: MARY> <instrument: CAR>

(28)      [+Pass] RUN OVER <patient: MARY> <instrument: CAR> <agent: X>

(29)      [-Pass] RUN OVER <patient: MARY> <instrument: CAR>

(30)      [-Pass] RUN OVER <patient: MARY> <instrument: CAR> <agent: X>

(31)      [-Pass] RUN OVER <patient: MARY> <instrument: CAR> <agent: [+Human]>

Compare (29), (30), and (31). In all three the accusative Case is assigned to MARY. No Case is assigned to CAR or to the agent argument. By the EPP, CAR must raise in (29). What about (30) and (31). The theta hierarchy also plays a role here. The theta-role highest on the hierarchy is raised and adjoined to TP where it is assigned the nominative Case. However, a small problem arises. 'agent:X' implies an agent but one not assigned a phonetic content. The grammar of English demands a phonetic subject (with certain predictable exceptions to be covered in more advanced syntax):

(32) *Ran over Mary with a car.

(32) Fails. In (29) CAR is raised and it is phonetic:

(33) A car ran over Mary.

The instrument in (30) cannot raise in the presence of an agent. When the instrument is raised, no agent is implied. There is no syntactic solution for (30). The logical structure for (30) is possible only in the passive when MARY is raised (22). We proposed above that 'agent:[+Human]' is spelled out as someone or somebody. In (31) the agent is raised since it is higher in the hierarchy than the instrument:

(34) [+Human] RUN-OVER-[+Past] MARY CAR.

(35) Someone ran over Mary with a car.

If the instrument is raised, we get the following ungrammatical sentence:

(33) *A car ran over Mary with someone.

Of course, a car and someone could have run over Mary (another reading where someone is concomitve), but this reading makes no sense, pragmatically.

Now consider the interrogative passive (see also 322 Questions):

(31). Was the potato eaten by John.

(32). Were stogies smoked by John?

Q adjoined to T, Pass lowers to V, and the secondary complement is raised (the following informal processes show empty categories which are not introduced until the final stage; they are shown here for convenience):

(33). [+empty] JOHN T+Q R Asp Vce SMOKE+Pass STOGIES+Acc [trace].

Agreement copies the feature of the subject onto T. (This version of html does not make subindexation possible. We will represent it in bold teletype font.):

(34). [+empty] JOHNi Ti+Q SMOKE+Pass STOGIES+Acc [trace].

Ti+Q is raised since Q is strong (it is adjoined to T):

(35). Ti+Q JOHN i [trace] SMOKE+Pass STOGIES+Acc [trace].

Since T is bound and it has no host, the dummy verb be is inserted to function as the host. T immediately c-commands Pass:

(36). {BE}+Ti+Q JOHN i [trace] SMOKE+Pass STOGIES+Acc [trace].

In the second pass to the lexicon {BE}+T[+Past]+[-Pers]+[-Pl] is spelled out as was.

Although Q is not assigned a phonetic segment, it is spelled out as a sentence contour--the question contour. The question mark at the end of the sentence represent the question contour:

(37) Are stogies smoked by John?

Since T still needs a host, {be} is inserted to function as a host, as T chain-governs [+Pass].

To return to course outline Click here.