Download as pdf or txt
Download as pdf or txt
You are on page 1of 17

Scope and Ambiguity in Expanded Categorial Semantics

Gary Hardegree
Department of Philosophy University of Massachusetts Amherst, MA 01003 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. Introduction......................................................................................................................................2 Ambiguity ........................................................................................................................................2 A More Complex Example ..............................................................................................................3 A Similar Example...........................................................................................................................6 A More Complex Example ..............................................................................................................8 Restrictive versus Non-Restrictive Adjectives ..............................................................................11 Adjectives Revisited ......................................................................................................................12 Initial Proposal ...............................................................................................................................13 Examples........................................................................................................................................14 Conclusion Semantic Scope; Compositional Ambiguity............................................................17

Hardegree, Scope and Ambiguity

page 2 of 17

1.

Introduction

In the present chapter, we discuss the notions of scope and scope-ambiguity, which play a central role in logic and semantics.

2.

Ambiguity

The traditional view is that ambiguity comes in two varieties lexical ambiguity, and structural ambiguity.

1.

Lexical Ambiguity

In instances of lexical ambiguity, a single morpheme has two or more entries in the lexicon; alternatively, the same surface form (spelling/pronunciation) corresponds to two or more morphemes. A prominent logical example is and which has two different meanings, as in: Jay respects Kay and Elle Jay is between Kay and Elle For example, the first one means that Jay respects Kay and Jay also respects Elle, but the second one does not mean that Jay is between Kay and Jay is between Elle, which is nonsense. Another example of lexical ambiguity is the suffix ed, which is used to mark passive voice but also to mark past tense.

2.

Structural Ambiguity

When three or more phrases are combined, the order of combination may be important to the content of the resulting compound phrase. For example, the following phrase two plus three times four admits two different analyses.

two

plus three times four two plus

times three

four

Structural ambiguity is also referred to as scope-ambiguity. In the first reading, plus has wider scope than times, whereas in the second reading times has wider scope than plus. Mathematics and logic uses parentheses to convey structure/scope. For example, we have the following expressions. 2 + (3 4) (2 + 3) 4 Sometimes the structural ambiguity is harmless, in the sense that the content is unaffected by the order of computation. For example,

Hardegree, Scope and Ambiguity

page 3 of 17

two plus three plus four computes to the same value regardless of how parentheses are inserted. Many times, however, a structural ambiguity is significant. For example, 2+(34)=14, but (2+3)4=20. Some phrases are both lexically-ambiguous and structurally-ambiguous, including the following one. poor violin player On the one hand, poor has three relevant lexical entries [poor versus rich, poor versus good, poor(ly) versus well]. On the other hand, it is also structurally ambiguous according to whether the violin player is poor or the violin is poor.

3.

A More Complex Example

The following example from elementary logic is a key example of a sentence that involves structural/scope ambiguity. every man respects some woman This plausibly has a standard SVO form, which is parsed as follows.

every

man

respects some woman

The corresponding semantic-tree goes as follows.


x { Mx y { Wy & Rxy } }

Q1 x { Mx Qx }

x1 y { Wy & Rxy }

P0 Q1 x { Px Qx } every1

x0Mx man

y2 x1 Rxy respects

Q2 y { Wy & Qy }

P0 Q2 y { Py & Qy } some2

y0Wy woman

However, in addition to this reading according to which every man has wide scope, there is another (less obvious) reading according to which some woman has wide scope, in which case the following is its translation. y { Wy & x { Mx Rxy } } In this connection, notice the following series.

Hardegree, Scope and Ambiguity

page 4 of 17

every man respects some woman some woman is respected by every man there is some woman who is respected by every man There are three accounts of this ambiguity.

1.

Logical-Form Ambiguity

According to this account, which is the more or less standard "logical" account of quantifierscope, in order to obtain the logical form of the above sentence, both quantifier phrases are raised to their logical position, which is at the front of their containing clause. Since there are two landing positions for the quantifier phrases, there are correspondingly two logical forms, given roughly as follows.

every man1
some woman2 e1 respects e2

some woman2
every man1 e1 respects e2

Here, e1 is the trace of every man and e2 is the trace of some woman. One must further postulate that the two QPs are lowered back into their respective trace positions in the final phonetic form.

2.

Structural Ambiguity

According to this account, the relative scopes of every man and some woman are indicated in the underlying tree-structure. In particular, in addition to the tree-structure given above, the following structure is also admissible.
y { Wy & x { Mx Rxy } }

y2x{Mx Rxy}

Q2 y {Wy & Qy}

Q1x{MxQx}

y2x1Rxy respects

P0Q2y{Py&Qy} some2

x0Wx woman

P0Q1x{PxQx} every1

x0Mx man

According to this account, we first combine every1 man and respects, which results in an accusative predicate [D2S], and then combine the resulting expression with some2 woman, which takes an accusative predicate and delivers a sentence.

Hardegree, Scope and Ambiguity

page 5 of 17

3.

Compositional Ambiguity

What is perhaps surprising is that one can achieve the same semantic result without adjusting the tree, but rather noticing that one of the compositions admits an alternative rendering, as seen in the following calculation.1
y { Wy & x { Mx Rxy } }

Q1 x { Mx Qx }

1 y { Wy & (x1Rxy) }

P0Q1x{PxQx} every1

x0Mx man

y2 x1 Rxy respects

Q2 y { Wy & Qy }

P0Q2y{Py&Qy} some2

y0Wy woman

The following is the key derivation, which is an application of the general inference principle we call "inflection". 1.
(1) (2) (3) (4) (5)

respects + some2 woman


y2 x1 Rxy Q2 y { Wy & Qy } 1 y2 1(x1 Rxy) y2 ([xRxy]) [Q2 y { Wy & Qy }] y2 ([xRxy]) [Q2 y { Wy & Qy }][y([xRxy])]2 y { Wy & [y([xRxy]y } y {Wy & (xRxy) } S 123 D2(D1S) (D2S)S (D1S)S D2S 1 2 3 13 Pr Pr As 1,4,Trans CC 2,4,O CC C C [(D1S)S]S 12 3,5,I

(6)

1 y { Wy & (xRxy) }

Note that, in order to save space, we pre-inflect every as nominative, and we pre-inflect some as accusative.

Hardegree, Scope and Ambiguity

page 6 of 17

4.

A Similar Example

The following sentence is another example from elementary logic that contains a scope ambiguity. every person is not virtuous The two admissible translations are given as follows. (t1) (t2) x{Px ;Vx} ;x{Px Vx} no person is virtuous not every person is virtuous

By way of explaining this ambiguity, there are a number of accounts. According to a fairly standard account, the scope of not is ambiguous. Specifically, either its scope is the whole sentence, or its scope is somewhere inside the VP is not virtuous. The following are the corresponding parsings.

not
every person is virtuous

every person virtuous is not

In the first case, we must further hypothesize that, in the final phonetic form, not is moved from its logical position at the head of the sentence to its surface position after the verb. In the second case, we must explain how is combines with not to form a phrase that combines with an adjective to form a VP. Expanded categorial grammar deals with the ambiguity in a different manner. First, we propose that the underlying grammatical structure is given as follows.2

every1

person is not

virtuous

The narrow-scope reading of not is then obtained by the following semantic calculation, where we treat virtuous as a bare-adjective.

Note that, in order to save space, we pre-inflect every to be nominative.

Hardegree, Scope and Ambiguity

page 7 of 17

x { Px ;Vx }

Q1x{Px Qx}

x1;Vx

P0Q1x{PxQx} every1

x0Px person P0{P1} is

P0x1;Px

x0Vx virtuous S;S not

The only non-standard maneuver is the composition of is and not, which is underwritten by the following derivation, which is an instance of transitivity. 1.
(1) (2) (3) (4) (5) (6) (7) (8) (9)

is + not
S;S P0{P1} P0 x1 P1 P1(x1) Px ;Px x1 ;Px P0 x1 ;Px S D1S (D0S)(D1S) 1234 123 12 SS (D0S)(D1S) D0S D1 D1S S 1 2 3 4 23 234 Pr Pr As As 2,3,O 4,5,O CC 1,6,O 4,7,I 3,8,I

The wide-scope reading of not is obtained, using the very same syntactic tree, as follows.
; x { Px Vx }

Q1x{PxQx}

1;(V)

P0Q1x{} every1

x0Px person

P01;(P)

x0Vx virtuous

P0{P1} is

S;S not

Once again, the only non-standard maneuver is the composition of is and not, but this time it yields a quite different expression, in accordance with the following derivation.

Hardegree, Scope and Ambiguity

page 8 of 17

2.
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)

is + not
S;S P0{P1} P0 1 x1 P1 P0(x0) Px x1Px P1 1(P1) (P) ;(P) 1 ;(P) P0 1 ;(P) S [(D1S)S]S (D0S){[(D1S)S]S} 1432 132 12 S 432 D1S 32 SS (D0S)(D1S) D0S (D1S)S D1 D1S S 1 2 3 4 5 23 325 Pr Pr As As As 2,3,@O 3,6,@O CC 5,7,@I def P1 4,8,O CC 1,9,O 4,10,I 3,11,I

5.

A More Complex Example


The following sentence is yet another example of quantifier scope-ambiguity. every friend of every virtuous person is virtuous

The word friend is an example of what we call a genitive noun,3 which has a type very similar to the type of transitive verbs the difference being the case-markings. type(friend) friend = = D6(D0S) y6x0Fxy [x is a friend of y]

In other words, friend takes a genitive-marked name and delivers a common noun. English marks genitive using apostrophe-s and also by of, which are analyzed as follows.4 's = of = x{x6}

Next, the following analysis is plausible.

3 4

See Chapter Relational Nouns and Prepositions for a more detailed account of genitive nouns. It is actually a little more complicated in English, since expressions like friend of mine appear to be twice-inflected. We propose that the word mine has secondary genitive-inflection, encoded by 5. Note in particular the difference between my/mine, her/hers, your/yours, their/theirs, our/ours. Usually, however, secondary-genitive is pronounced just like genitive, as in friend of Kay's, or not pronounced at all.

Hardegree, Scope and Ambiguity

page 9 of 17

x{ y{VyFxy} Vx }

Q1x{y{VyFxy}Qx}

x1Vx

P0Q1x{PxQx} every1

x0y{VyFxy}

P0{P1} is

x0Vx virtuous

y6x0Fxy friend x{x6} of

Q6y{VyQy}

Qy{VyQy}

P0Qy{PyQy} every

x0Vx virtuous person

In this analysis, we let the domain be persons, so we do not separately translate person. We also treat the second occurrence of virtuous as a bare-adjective, which has the same type as a common-noun. The upshot of this analysis is that the sentence says that, if a person is a friend of every virtuous person, then that person is virtuous. This is not the most plausible reading of the sentence. Indeed, if we change the word friend to child or wife, then its plausibility drops significantly, for then its says that if you are a child/wife of every virtuous person, which is highly unlikely, then you are virtuous. A more plausible reading reverses the respective scopes of the two quantifier phrases. Using the structural account of scope, we have the following alternative tree.
y{ Vy x{FxyVx} }

Q1y{ Vy x{FxyQx} }

x1Vx

y6Q1x{FxyQx}

Q6y{VyQy}

P0{P1} is

x0Vx virtuous

P0Q1x{PxQx} every1

y6x0Fxy friend

x{x6} of

Qy{VyQy}

P0Qy{PyQy} x0Vx every virtuous person

According to this reading, if a person is virtuous, then every friend of that person is virtuous. The key compositions are given by the following derivations.

Hardegree, Scope and Ambiguity

page 10 of 17

1.
(1) (2) (3) (4) (5) (6)

every + friend
P0Q1x{PxQx} y6x0Fxy y6 x0Fxy [P0Q1x{PxQx}]x0Fxy Q1x{FxyQx} y6Q1x{FxyQx} D6[(DS)S] 12 (D0S)[(DS)S] D6(D0S) D6 D0S (DS)S 1 2 3 23 123 Pr Pr As 2,3,@O 1,4,@O CC,@C 3,5,@I

2.
(1) (2) (3) (4) (5)

every friend + of every virtuous person


y6Q1x{FxyQx} Q6y{VyQy} Q1 y6x{FxyQx} [Q6y{VyQy}]y6x{FxyQx} y{ Vy x{FxyQx} } D6[(D1S)S] (D6S)S D1S D6S S 1 2 3 13 213 Pr Pr As 1,3,@O2 2,4,@O CC,@C

We can also obtain this reading using the original tree, but employing a different composition of
friend and of every virtuous person, as follows. x{ y{VyFxy} Vx }

Q1y{Vyx{FxyQx}}

x1Vx

P0Q1x{PxQx} 0y{Vy(xFxy)} every1 y6x0Fxy friend x{x6} of Q6y{VyQy}

x1{x0} is

x0Vx virtuous

Qy{VyQy}

P0Qy{PyQy} every

x0Vx virtuous person

The two novel compositions are achieved by the following derivations.

Hardegree, Scope and Ambiguity

page 11 of 17

3.
(1) (2) (3) (4) (5) (6) (7)

friend + of every virtuous person


y6x0Fxy Q6y{VyQy} 0 y60(x0Fxy) y6(xFxy) [Q6y{VyQy}]y6(xFxy) y{ Vy (xFxy) } 0y{ Vy(xFxy) } ((D0S)S)S 21 S 213 D6(D0S) (D6S)S (D0S)S D6S 1 2 3 13 Pr Pr As 1,3,@O CC 2,4,@O @C 3,6,@I

4.
(1) (2) (3) (4) (5)

every1 + friend of every virtuous person


P0Q1x{PxQx} 0y{Vy(xFxy)} Q1 P0x{PxQx} [0y{Vy(xFxy)}]P0x{PxQx} y{ Vy [Px{PxQx}](xFxy) } y{ Vy x{ FxyQx } } (D0S)[(D1S)S] ((D0S)S)S D1S (D0S)S S 1 2 3 13 213 Pr Pr As 1,3,@O2 2,4,@O CC,@C @C (D1S)S 21 3,5,@I

(6)

@Q1y{ Vy x{ FxyQx } }

6.

Restrictive versus Non-Restrictive Adjectives

Another application of scope arises in connection with adjectives and relative clauses. We have already discussed restrictive relative clauses, which we treat as adjectives. In the next few sections, we discuss non-restrictive relative clauses and non-restrictive adjectives. The following pairs illustrate the difference between restrictive and non-restrictive relative clauses. (1) (2) (3) (4) the woman-who-is-tall the woman who is tall there is exactly one woman-who-is-tall there is exactly one woman who is tall [restrictive] [non-restrictive] [restrictive] [non-restrictive]

The punctuation has been exaggerated to accentuate the difference. In the two restrictive examples, woman who is tall forms a constituent, but in the non-restrictive examples, there is a definite logical/phonological break between woman and who. Semantically speaking, in the first case, there may be many women in the relevant situation, but exactly one of them is tall. In the second case, there is exactly one woman in the situation, and she is tall. The overall syntactic structures are given as follows.

Hardegree, Scope and Ambiguity

page 12 of 17

the woman-who-is-tall

there is exactly one woman-who-is-tall

the woman the woman, who is tall who is tall

there is exactly one woman there is exactly one woman, who is tall who is tall

who is tall the woman there is exactly one woman

who is tall

Notice that, in the first pair, the scope of who is tall is narrow, but in the second pair, its scope is wide.

7.

Adjectives Revisited

Or current proposal is that restrictive relative clauses are a species of adjective.5 On the other hand, according to our current proposal, adjectives are modifiers, where a modifier is a functor of type , for some type . An adjective is, in particular, a noun-modifier. However, since there are two kinds of nouns proper-nouns and common-nouns there are in principle two kinds of adjectives. On the one hand, an ordinary adjective is a common-noun modifier, which has the following type. type(CN-Adj) = CC (D0S)(D0S)

What then is a proper-noun adjective? We propose that a proper-noun adjective is a functor of the following type. type(PN-Adj) = (DS)(DS)

Notice that the latter type is isomorphic to the earlier type, by which we mean they are identical except for case-inflection. Notice, however, that the proposed type does not look exactly like a proper-noun modifier. It accordingly might be useful to transform it as follows, in accordance with categorial logic. (DS)(DS) D[(DS)S] DQP

Adjectives come in two guises bare-adjectives [type D0S] and modifier-adjectives [type (D0S)(D0S)]. Unless marked by the term bare, an adjective is understood to be a modifier-adjective.

Hardegree, Scope and Ambiguity

page 13 of 17

In this guise, the functor takes a proper-noun phrase as input and delivers a QP as output, which is dual to a proper-noun phrase. See later examples for how this works in practice. The obvious remaining question is whether there are any proper-noun adjectives. The following seem to be plausible examples. evil Bart big bad John curious George Here, it is fairly clear that the adjectives modify proper-nouns. The following examples are more interesting, since they are ambiguous. the industrious Chinese the stupid president The first one is well-known, but is perhaps a little bit out of date.6 It also involves a plural term, which we prefer to avoid when we can. So let us concentrate on the second example, which is singular and more up to date.7 It is evident that it is ambiguous between the following readings involving relative clauses in one case, restrictive; in the other case, non-restrictive. the president-who-is-stupid the president, who is stupid For example, in the first case, we are talking about a collection of presidents, and we are referring to the stupid one. On the other hand, in the second case, we are referring to exactly one president, who we are parenthetically remarking is stupid. We next note that our earlier examples of proper-noun modifiers can also be paraphrased using non-restrictive relative clauses. Bart (who is evil) John (who is big and bad) George (who is curious)

8.

Initial Proposal
We initially propose the following parallel.8 common-noun adjectives proper-noun adjectives restrictive relative clauses non-restrictive relative clauses

In particular, we propose the following categorial identities.

It is not that the Chinese are no longer thought to be industrious. Rather, it is no longer considered politically-correct to stereotype groups, even if the stereotype is positive. 7 This too is not up-to-date, as of January 2009, when Barack Obama became president of the U.S. Perhaps, we can combine the two examples into the industrious president. 8 In a later chapter, Anaphoric Uses of Relative Pronouns, we reconsider this proposal.

Hardegree, Scope and Ambiguity

page 14 of 17

type(NRCF) type(RRCF)

= =

VP CN-Adj (DS)[(D0S)(D0S)] VP PN-Adj (DS)[(DS)(DS)]

We further propose a non-restriction operator [nr] which converts a standard modifieradjective of the form P0 x0 ( Px & Ax ) (D0S)(D0S) into its non-restrictive counterpart: P x ( Px & Ax ) (DS)(DS)

The operator's lambda-translation is given as follows, where is an adjective-variable. P x (P0)(x0) The following series of lambda-conversion steps shows how [nr] works on an intersective adjective; it basically removes the zero-inflections. [ P x (P0)(x0) ] P0 x0 { Px & Ax } P x [P0 x0 { Px & Ax }](P0)(x0) P x [x0 { Px & Ax }](x0) P x { Px & Ax } = = =

Alternatively, we can take advantage of generalized lambda-abstraction, and write the following more complex, but more obvious, formulation. nr = {P0x0}{Px}

9.
1.

Examples
Jay, who respects no man, respects Kay
RJK & ;y{My & RJy}

P1{PJ & ;y{My & RJy}}

x1RxK respects Kay2

P{PJ & ;y{My & RJy}}

[+1]

Px{Px & ;y{My & Rxy}}

Jay {P0x0}{Px} [nr] P0x0{Px & ;y{My & Rxy}} who respects no man

Hardegree, Scope and Ambiguity

page 15 of 17

2.

the greatest-book, which Ptolemy wrote, is Almagest 9


xGx " A & W[P,xGx]

P1 { P[xGx] & W[P,xGx] }

x1{x"A} is Almagest

P { P[xGx] & W[P,xGx] }

x1{x"R} [+1]

xGx the greatest-book

P x { Px & W[P,x] }

{P0x0}{Px} [nr]

P0x0{Px & W[P,x]}

Q2P0x0{Px & Qx} which2


P1

y2W[P,y]

Ptolemy1

y2x1W[x,y] wrote

3.

curious George is a monkey


MG & CG

P1{PG & CG}

x0Mx is a monkey x{x1} [+1]

P{PG & CG}

Px{Px & Cx}

George {P0x0}{Px} [nr] P0x0{Px & Cx} curious (mod)

Almagest is the Arabic title for this (Alexandrian) Greek work; it means the greatest.

Hardegree, Scope and Ambiguity

page 16 of 17

4.

the (stupid) president respects no one


The following is the semantically most natural rendering of this sentence.
;y R[xPx, y] & S[x Px]

Q1 { Q[x Px] & S[x Px] }

x1 ;yRxy respects no one x{x1} [+1]

Q { Q[x Px] & S[x Px] }

xPx

Qx{Qx & Sx}

P0 x Px the

x0 Px president

{P0x0}{Px} [nr]

Q0x0{Qx & Sx} stupid (mod)

The problem is that stupid is syntactically misplaced. We can plausibly regard this as a purely syntactic issue. For example, the corresponding French expression would be in correct order. In this case, we postulate that the order is appropriately modified by the phonetic module. If one is not entirely satisfied with this solution, one can alternatively propose the following semantic derivation.
;y R[xPx, y] & S[x Px]

Q1 { Q[x Px] & S[x Px] }

x1 ;yRxy respects no one

Q { Q[x Px] & S[x Px] }

x{x1} [+1]

P0 x Px the

Q { Q[(P0)] & S[(P0)] }

Q x { Qx & Sx }

x0 Px president

{P0x0}{Px} [nr]

Q0 x0 { Qx & Sx } stupid (mod)

The novel compositions are underwritten by the following derivations.

Hardegree, Scope and Ambiguity

page 17 of 17

1.
(1) (2) (3) (4) (5) (6)

stupid + president
Q x { Qx & Sx } x0 Px [= P0] (P0) Q { Q[(P0)] & S[(P0)] } Q{ Q[(P0)] & S[(P0)] } (DS)(DS) D0S (D0S)D D (DS)S [(D0S)D][(DS)S] 1 2 3 23 123 12 Pr Pr As 2,4,O 1,6,O2 3,5,I

2.
(1) (2) (3) (4) (5)

the + stupid president


Q { Q[(P0)] & S[(P0)] } P0 xPx [ Q { Q[(P0)] & S[(P0)] }] P0 x Px Q { Q[[P0 x Px]P0] & S[[P0xPx]P0] } Q { Q[xPx] & S[xPx] } [(D0S)D][(DS)S] (D0S)D (D0S)D 1 2 12 Pr Pr 1,2,O C C

10.

Conclusion Semantic Scope; Compositional Ambiguity


We draw two conclusions from our deliberations above. (1) The scope of a functor can be captured semantically in a manner that does not show up in the parse tree.

For example, in the last tree, there is no way syntactically to determine that stupid has wide scope. Nevertheless, its meaning manages to percolate up the tree to become semantically dominant. We propose to call this semantic scope. (2) In addition to lexical ambiguity and structural ambiguity, there is also compositional ambiguity.

We have seen three examples of this. In the last example, the two semantic trees have exactly the same structure, and they have exactly the same lexical starting points, and yet they do not produce the same semantic value at the top node. This is, indeed, a direct consequence of our account of semantic-composition. We propose that and combine to form precisely if follows from and in the categorial logic. Generally speaking in logic, more than one formula follows from a given set of premises.

You might also like