Spring term, 2020 Ling 5201 Syntax I 1: Valence, rules and proof - - PowerPoint PPT Presentation

Spring term, 2020 Ling 5201 Syntax I 1: Valence, rules and proof

Robert Levine

Ohio State University levine.1@osu.edu

Robert Levine 2020 5201 (Syntax 1) 1 / 31

The first question: Where does our notion ‘syntactic structure’ come from?

◮ In any course you’ve ever taken where syntax was covered as at least

part of the course, you undoubtedly heard syntactic structure used throughout the course,

◮ very possibly without any explanation of why we associate the notions

‘syntax’ and ‘structure’ so automatically.

◮ It’s as though it’s so obvious that sentences—the linguistic objects

that syntax is about—have some kind of structure that the reasoning to that conclusion needn’t ever be made explicit.

◮ But we can’t have that. We are obliged to make clear what the notion

• f structure is, and why syntax has come to become associated, by

strong default, with this notion.

◮ What motivates our belief that sentences have a particular kind of

structure, and what is that kind?

Robert Levine 2020 5201 (Syntax 1) 2 / 31

Some history

◮ Modern grammatical theory has a direct line of descent from the

anthropologicaltradition in linguistics inaugurated by Franz Boas and his students,

◮ who rejected the evolutionary view of language held by the European

19th century historical comparativists,

◮ replacing it with an approach to analysis whose core premise was that

no language could be analytically understood unless it was described according to its own particular patterns and internal organization.

◮ In practical terms, this meant that the grammar of a language

consisted of a set of patterns defined on the basis of the distribution

• f particular word sequences, allowing the linguist to identify classes of

items with parallel distribution.

◮ Members of such classes were regarded as the combinatorial units

whose distribution was determined and controled by the rules of the grammar.

Robert Levine 2020 5201 (Syntax 1) 3 / 31

History, cont’d

◮ This view of the grammar was heavily influenced by the study of morphologically

complex languages by Boas, his students and their students (and by others strongly influenced by these).

◮ So we find, in the 1940s, in the era known as American Structuralism, analyses of

words from native North American languages such as (1), for the Totonac word kilila·pa·ˇ ski·qu·t (‘my necessity of loving them reciprocally’): (1)

kilila·pa·ˇ ski·qu·t lila·pa·ˇ ski·qu·t t la·pa·ˇ ski·qu· qu· la·pa·ˇ ski· pa·ˇ ski· la· li ki

◮ I’m not reinterpreting the source here: this is exactly the picture that Eugene Nida,

in his great summary of Structuralist methods of morphological analysis Morphology, provides for the Tontonac word.

Robert Levine 2020 5201 (Syntax 1) 4 / 31

History, cont’d

◮ What exactly do such graphic object signify? A graph can have any number of formal interpretation. ◮ To the American Structuralists, such a graph signified primarily the order of analysis. You’re working your way in from the outside, segmenting off meaning-bearing pieces based

• n paradigmatic information,

◮ with li. . . t a circumfix. ◮ So (1) represents a history of analysis—the breakdown of the word as carried out according to the distributional properties of its parts as recoverable from Totponac morphological paradigms. ◮ The Structuralists extended this approach to analytic representations to syntax as well, giving us a breakdown for The King of England opened Parliament as (2)

the King of England opened Parliament

• pened Parliament

Parliament

• pened

the King of England King of England

• f England

England

• f

King the

◮ Does this look like the kind of syntactic tree you’re used to seeing? WHAT’S MISSING??

Robert Levine 2020 5201 (Syntax 1) 5 / 31

Some data

◮ Consider (3):

(3) John put this book on that table. (4) Here’s a typical elementary-syntax-level representation of the structure of (3):

(5)

S VP PP NP N table Det that P

• n

NP N book Det this V put NP John ◮ The interpretation of this tree, compared to (2), differs in two related respects.

◮ First, the nodes of the tree are labeled with certain symbols that, as we know,

identify the syntactic category or type of the word string that can be linked by pathways back to each such node.

Robert Levine 2020 5201 (Syntax 1) 6 / 31

Data, cont’d.

◮

◮ Second, the information in (5) is standarly taken to represent properties

• f the sentence. That is,

◮ the sentence is composed of and NP and a VP, and the VP in turn

comprises three subunits, a verb, a N(oun)P(hrase) and a P(repositional)P(hrase), and so on.

◮ These properties are given as facts about the syntactic form of the

sentence in (3).

◮ In effect, this representation is saying that a determiner and a NP

combine to form an NP this book in one case and that table in another, that the P of combines with the NP that table to form a PP, that this NP and PP combine with the verb put to form a VP, etc.

◮ Two questions:

◮ What justifies this picture of syntactic representations? ◮ Is there another picture that’s in principle possible? Robert Levine 2020 5201 (Syntax 1) 7 / 31

Justifying tree structure

◮ There are three things you need to do to justify (5):

◮ You need to justify the existence of the labeled nodes themselves as representations

• f syntactic units;

◮ you need to justify the labels attached to the nodes; and ◮ you need to justify the existence of that hierarchical organization of nodes as part of

the syntactic representation for (5). ◮ Justifications for the existence of syntactic units, aka constituents, typically starts

with a comparison of (6) and (7) (6)

• a. John will put this book on that table.
• b. this book, John will put on that table

.

• c. that table, John will put this book on

.

• d. On that table, John will put this book

.

• e. and put this book on that table John most certainly will

. (7)

• a. *book on, John will put this

that table.

• b. *On that, John will put this book

table.

• c. *Book on that, John will put this

table.

• d. *Put this, John will

book on that table.

Robert Levine 2020 5201 (Syntax 1) 8 / 31

Justifying, cont’d.

◮ There are only certain substrings of (6a) that can appear in leftmost

position at the edge of a version of it which is identical to (6a) except that it’s missing that substring.

◮ The key idea is that in order to capture the distinction between the

good and the ill-formed examples, you need to be able to refer to some object which can occupy that left side slot,

◮ while at the same time ensuring that an object of that type is missing

from somewhere within rest of the sentence.

◮ But being able to refer to such an object presupposes that it exists as

a component of the representation.

◮ It follows that we have a strong plausibility argument for taking the

sentences in (6) to reflect the existence of syntactic constituents.

◮ But what should we call these objects?

Robert Levine 2020 5201 (Syntax 1) 9 / 31

Where do the type names come from?

◮ Let’s elaborate (6a) a bit and see what we can learn about what the essential

properties are of the left-edge constituent. (8)

• a. that strange photo of Chris, Robin will put
• n this desk.
• b. that photo of Chris, Robin will put
• n this desk.
• c. that strange photo, Robin will put
• n that desk.
• d. Strange photos of chris, Robin will put
• n this desk.
• e. strange photos, Robin will put
• n this desk.
• f. Photos of chris, Robin will put on this desk
• g. photos, Robin will put
• n this desk.

(9)

• a. *That strange (of Chris), Robin will put
• n this desk.
• b. *Of Chris, Robin will put
• n this desk

◮ Apparently, the one item you cannot omit here is photo(s), an N, when you are missing material directly following put. ◮ This constitutes evidence that one of the possible constituent types which can show up on the left edge of the sentence in such cases reflects something which is inherently nominal in character, ◮ hence meriting the description ‘noun phrase’ (NP).

Robert Levine 2020 5201 (Syntax 1) 10 / 31

Type names, cont’d.c

◮ We can then characterize all of the left-positioned strings in (8) NPs. ◮ Similar reasoning can be applied to (10):

(10) . . . and argue with Mary about politics, John definitely will .

◮ We will then be able identify the all of the following as VPs:

(11)

• a. argue
• b. argue with Mary
• c. argue about politics
• d. argue with Mary about politics.

◮ So we can use the optional vs. . essential distinction to identify the crucial

distributional property of different syntactic types in terms of the lexical category whose distribution each type mirrors:

◮ NPs reflect the distribution of nouns (cat(s)); VPs the distribution of verbs

(argue), etc.

Robert Levine 2020 5201 (Syntax 1) 11 / 31

Type names, cont’d.

◮ This mode of reasoning thus establishes the second of the three points

in question, which were:

◮ the identification of the nodes in trees as objects encoding the syntactic

unithood of the word strings under those nodes;

◮ the use of labels for these nodes of the form XP, where X is some

lexical category, and

◮ the existence of labeled nodes under other labeled nodes.

◮ It’s the last of these points which is arguably the most fundamental

issue in all of syntactic theorizing.

◮ To understand what’s at stake, we need to revisit (6a) in the light of

the first two points.

Robert Levine 2020 5201 (Syntax 1) 12 / 31

Why internal structure?

◮ The tree for (6a) is a slightly more elaborate version of (5): (12)

S VP PP NP N table Det that P

• n

NP N book Det this V put V will NP John

◮ Let’s agree that by the methods we’ve used, we can show that a V put, an NP this book and PP on this desk combine to give us a string of words that has the distributional properties required of a VP, ◮ and that when this string of words combines with the verb will and the NP John, the result is a sentence of English. ◮ That’s the information that’s present in the tree for John will put this book on that table. But, given that combining put, this book and on this desk form a VP, why do we need to have that information present when the VP and the NP John combine to form a sentence? ◮ In other words,why do we need to mark the internal structure of the sentence as in (12)?

Robert Levine 2020 5201 (Syntax 1) 13 / 31

Why internal structure, cont’d

◮ To ask this question is in effect to answer the question posed above: is there any

logically possible alternative to syntactic representations such as (12)?

◮ Suppose we said that, as per what we’ve already decided, a V put, an NP this book

and a PP on this desk combine to give a VP put a book on this desk, with no internal marking of these components.

◮ This VP combines with an NP John and with a V will to give an S, but again, with

no internal marking.

◮ On this approach, if two constituents X and Y combine to give a constituent A,

the string of words labeled by X and the string of words labeled by Y combine, and that result is labeled A, but there is no longer any labeling X, Y .

◮ Think about how a jigsaw puzzle gets put together. . . ◮ So why not this much simpler picture?

Robert Levine 2020 5201 (Syntax 1) 14 / 31

Why internal structure, cont’d

◮ It turns out that there is a single, straightforward reason for the near-universal

assumption of hierarchical phrase structure representation in the history of modern syntactic theory. . .

• . . . and we’ve already seen the main empirical evidence for it.

◮ Look again at the examples of extraction that we’ve been basing this discussion on:

(13) John will put this book on that table. (14)

• a. [np this book], John will put on that table

.

• b. [np that table], John will this book on

.

• c. [pp On that table], John will put this book

.

• d. and [vp put this book on that table] John most certainly will

.

◮ State a simple descriptive rule that will account for the relationship between (13)

and (14).

Robert Levine 2020 5201 (Syntax 1) 15 / 31

Why internal structure, cont’d

◮ Try the following:

(15) A phrasal constituent within a canonical sentence S may appear in a position immediately preceding S rather than its position within S.

◮ If you assume, as is virtually universally the case, that extraction constructions are

to be defined in terms of a mapping from a canonical, purely in situ sentence to an extraction version,

◮ then something along the lines of (15) is essentially inevitable. ◮ But the only way this definition of the possibilities can work is if constituency is

defined within the canonical sentence S to which the (15) applies!. . .

◮ . . . Right? ◮ Actually, wrong. ◮ But this assumption, which seems to compel agreement, turns out to be, in its

entirety, an artefact of a specific premis about how strings of words are to be assigned syntactic types,

◮ a premise encoding a view we can call syntax as construction.

Robert Levine 2020 5201 (Syntax 1) 16 / 31

Syntax as construction

◮ I’m going to constrast this approach to defining what sequences of words in a

language are actual sentences with something we can call syntax as deduction,

◮ an alternative whose roots go back to the 19th century, ◮ and which, I’ll try to show, is far more successful at handling a range of phenomena

that have always wound up being embarrassments for the constructive version of syntax.

◮ But for the time being, let’s just try to make the syntax-as-construction paradigm

more explicit.

◮ The core technical device in this approach is a statement which identifies a

category symbol on the left of an arrow and provides a string of one or more category symbols on the right,

◮ telling you, in effect, that a sequence of word strings each of which meets the

description of the successive symbols on the right is a string belonging to the syntactic type identified by the symbol on the left.

◮ E.g., a rule A → XY Z tells you that a word string belonging to type X followed

by a word string belonging to type Y followed by a string belonging to type Z is a string belonging to type A.

Robert Levine 2020 5201 (Syntax 1) 17 / 31

Rules and trees

◮ Here’s a practical illustration of this approach, showing why it’s constructive in nature. (1)

S VP PP NP N table Det that P

• n

NP N book Det this V put NP John

(16)

• a. S → NP VP
• b. VP → V NP PP
• c. PP → P NP
• d. NP → Det N

◮ To make sure that names such as John and Mary don’t appear with determiners immediately preceding them, we assume that they are one-word NPs. ◮ Each rule admits one or more nodes in the tree in (5) (potentially subject to certain restrictions about what kinds of types allow their string inhabitants to appear in left-edge initial position). ◮ Each such node is an hypothesis about the displaceability of the material below that node. ◮ The terminal nodes dominate individual words whose properties are identified in a lexicon, and include

◮ information about pronunciation ◮ about part of speech

◮ A string of terminal nodes licenses a string of words where for all n, the nth word in the string is assigned the nth lexical type.

Robert Levine 2020 5201 (Syntax 1) 18 / 31

Phrase structure rules, part 2

◮ Clearly, the rules in this toy grammar must be elaborated quite a bit in

• rder to achieve anything like realistic coverage:

(17)

• a. The cat slept.
• b. The cat chased a mouse.
• c. Fat cats slept.
• d. The book of poems fell on the floor.

◮ And this is just the beginning. ◮ To accommodate these, we need to replace the rules we’ve given with

something more like: (18)

• a. S → NP VP
• b. VP → V (NP) (PP)
• c. PP → P NP
• d. NP → (Det) (Adj) N (PP)

Robert Levine 2020 5201 (Syntax 1) 19 / 31

The lexicon

◮ The lexicon determines how the trees sanctioned by the rules can be

terminated: (19) John: john, NP table: table, N the: the, Det put: put, V lawyers: lawyers, N

• n:
• n,

P talked: talked, V discussed: discussed, V about: about, P various: various, Adj some:, some, Det . . . . . . . . .

◮ The rules given allow us to license the sentences in (20).

(20)

• a. The lawyers talked about various issues.
• b. John talked about some lawyers.
• c. The lawyers put some issues on the table.

◮ But is this enough?

Robert Levine 2020 5201 (Syntax 1) 20 / 31

The lexicon, cont’d

◮ The following tree is legal, given the rules (18) and lexicon (19), but the corresponding string is not: S VP V discussed NP N lawyers Adj various (21) *Various lawyers discussed. ◮ And it gets worse: (22)

• a. *The lawyers put.
• b. *Some crafty lawyers put the issue.
• c. Various lawyers talked.
• d. *Various lawyers talked the obscure issues.
• e. Various lawyers talked about the obscure issues.
• f. *Some lawyers discussed.
• g. Some lawyers discussed the issues.
• h. *Some lawyers discussed about the issues.

Robert Levine 2020 5201 (Syntax 1) 21 / 31

Valence

◮ Different lexical items have different combinatorial possibilities. ◮ talk combines with a PP headed by about. . . ◮ . . . whereas discuss combines with an NP:

(23)

• a. Various lawyers discussed the obscure issues.
• b. Some lawyers talked about the issues.

◮ and you can’t reverse the patterns in (23):

(24)

• a. *Various lawyers talked the obscure issues.
• b. *Some lawyers discussed about the issues.

◮ The combinatorial possibilities of a lexical item are called its valence.

Robert Levine 2020 5201 (Syntax 1) 22 / 31

Valence is not predictable. . .

◮ Do we need to treat valence as a primitive property of lexical items? Can’t we predict it from meaning? ◮ In a word, NO. (25)

• a. John ate (a steak)
• b. John devoured *(a steak).

(26)

• a. Mary demanded a raise
• b. *Mary (authoritatively) asked a raise.

(27)

• a. *Mary demanded for a raise
• b. Mary (authoritatively) asked for a raise.

(28)

• a. The prosecutor

8 < : ∗charged ∗indicted accused 9 = ; John of forgery

• b. The prosecutor

8 < : ∗charged ∗accused indicted 9 = ; John for forgery.

• c. The prosecutor

8 < : ∗accused ∗indicted charged 9 = ; John with forgery

Robert Levine 2020 5201 (Syntax 1) 23 / 31

Valence is not predictable. . .

(29) Mary  hopes aspires ff to win the Booker Prize. (30) Mary  hopes *aspires ff that she will win the Booker Prize. (31) Mary  longs aspires ff to win the Booker Prize. (32) Mary  hopes *aspires ff for great things. (33) Mary  *hopes aspires ff to great things. (34) Mary  longs *aspires ff *for the Booker Prize. ◮ Both hope and aspire take irrealis properties (or propositions) as arguments, as is consonant with their semantics. ◮ We associate irrealis forms with nonfinite tense and realis forms with finite tense. . . ◮ But you can do irrealis in finite contexts in a number of ways, including modal auxiliaries (will, could, etc.) ◮ The point is that if aspire could take a finite clause argument, there would be a way to make that clause irrealis—but it just can’t. ◮ The bottom line: when it comes to valence, syntax RULES

Robert Levine 2020 5201 (Syntax 1) 24 / 31

. . . hence, valence must be lexically listed

◮ We start by taking lexical entries to specify pronunciation, part of

speech and valence, along the following lines:

◮ where the notation

X Y Z etc. means only that the lexical item in question appears followed by a string of words which can be parsed as a substring of category X, which is then followed by a string of words of category Y, etc.

(35) put, V, NP PP discuss, V, NP talk, V, PP charge, V, PP[with] indict, V, PP[for] accuse, V, PP[of] . . . . . . . . .

◮ But such entries have both too much and too little information:

◮ they lack information about meaning; ◮ they give you information which, in an important sense, you don’t need.

◮ Taking the last point first: if we know what a word combines with to

yield a VP,

◮ then it doesn’t really matter what part of speech we assign it to.

Robert Levine 2020 5201 (Syntax 1) 25 / 31

Valence reconsidered

◮ Let’s use a presentation of categories along these lines which displays

more explicitly the direction of combination.

◮ We have an entry for put of the form

(36) put; put; VP/PP/NP

◮ i.e., the verb put

◮ has a certain pronunciation, which we abbreviate as put, ◮ has a semantic interpretation, which we abbreviate as put, ◮ and combines to its right first with an NP (to yield a string such as put

the book)

◮ and then a PP (possibly yielding put the book on the table).

◮ Following the line just suggested, we can revise our valence entries as

follows: (37) put; put; VP/PP/NP discuss; discuss; VP/NP talk; talk; VP/PP charged; charge; VP/PP[with] accused; accuse; VP/PP[of] indicted; indict; VP/PP[for]

Robert Levine 2020 5201 (Syntax 1) 26 / 31

Reconsidering phrase structure: some problems that won’t go away

◮ At this point, I’ve sketched the bare bones of the kind of thinking that

has essentially dominated syntax almost completely for close more than sixty years.

◮ That kind of dominance tends to create the illusion of overwhelming

empirical success, attaining maximum explanatory insight with the minimum technical machinery necessary.

◮ But it is an illusion, nonetheless.

Robert Levine 2020 5201 (Syntax 1) 27 / 31

Problems, cont’d

◮ Consider the following kinds of patterns in English.

(38)

• a. John [VPfin ate the cake] and [VPfin seemed to like it].
• b. John wanted to [VPbse walk] and [VPbse talk].
• c. John [V likes] and [V admires] Mary.
• d. [NP John ] and [NP Mary] went to a movie.
• e. John drove [PP out of the garage ] and [PP onto the street]

◮ These instances of coordination all reflect combinations of constituents that

we recognize from the extraction diagnostic.

◮ But if we assume that these are the only combinatoric units in the grammar,

how do we account for the examples in (39)? (39) Bill put [the books on the table] and [the vase on the shelf].

◮

The sequences in brackets are coordinates (as we’ll be able to show in due course),

◮

but they do not satisfy the core diagnostic for constituency.

Robert Levine 2020 5201 (Syntax 1) 28 / 31

Problems, cont’d

◮ Again, consider the examples in (40) and (41):

(40) John put books on the table, and Mary, on the shelf. (41) John can fit more books on the table than Bill can on the shelf.

◮ In (40), we appear to be coordinating a sentence John put books on the table with

something that isn’t even a constituent, let alone a sentence.

◮ In (41), we seem to have a sequence (Bill on the shelf )

◮ which is missing material (fit more books) that, again, no one would argue is a PS

constituent,

◮ while *John can on the shelf, on its own in isolation, is not a sentence.

◮ The problem is not that the concept of ‘constituent’, as a combinatorial unit, is

misguided,

◮ but rather that treating the syntactic respresentation of a sentence as a reflection

• f constructive rules of the kind we’ve seen makes it impossible to capture the

critical notion of a syntactic type from which material of another (possibly different) type is missing,

◮ just the very situation that we see illustrated in these examples of coordination and

ellipsis.

Robert Levine 2020 5201 (Syntax 1) 29 / 31

A different view of categories. . .

◮ The moral of the story is that something different from the PS

rule-based grammar is needed,

◮ one which starts from the premise that a string of words may be an

instance of a constituent belonging to a type Y that is missing material of type X.

◮ For example: our PS rule for sentences S → NP VP tells us that a

sentence is formed by combining an NP string to the left of a VP string.

◮ But we can reframe this combinatoric relationship by identifying VP

string as strings of words looking for an NP to the left to form a sentence,

◮ which we encode by replacing ‘VP’ with the type name NP\S:

something that will combine with an NP on the left to yield an S.

◮ A transitive verb such as discuss or criticize, i.e., a linguistic sign

which combines with an NP sign to its right to yield a VP, can be written VP/NP, or, unpacking the abbreviation VP, (NP\S)/NP.

Robert Levine 2020 5201 (Syntax 1) 30 / 31

A different view, cont’d.

◮ How does this approach change our view of syntactic structure? ◮ One one level, not all that much. Compare the trees:

(42)

S VP NP Bill V criticized NP Mary S NP\S NP Bill (NP\S)/NP criticized NP Mary

◮ Seemingly, all that has happened is the replacement of category

names based on parts of speech with category names based on combinatory possibilities, i.e., valence.

◮ But this reformulation leads directly to a foundationally different view

• f the way we think about the syntax of sentences,

◮ one in which sentences are in effect theorems of a kind of logical

deduction which is in a strict correspondence with standard logics of truth and inference.

Robert Levine 2020 5201 (Syntax 1) 31 / 31