The Development of X-bar theory

  1. What Came Before
  2. No notion of phrase in Antiquity because classical languages were morphologically complex
  1. enough to concentrate on with morphology
  2. main parts of sentences could be associated with words
  1. subject = word bearing nominative morphology
  2. subject word agrees with verb and modifying words
  1. a lot of word order freedom hides phrases
  2. American Structuralism
  3. Operationalism – only observed things can be part of science. BUT most of linguistics can’t be observed
  4. So everything must be based on what can be observed – i.e. phonetics à phonology à syntax
  5. The elements of each higher level must be based on elements of the lower levels to be rooted in observation
    phones à phonemes à morphemes à words à phrases à sentences
  6. How these are linked to each other is through the notion of distribution
    a phoneme is a set of phones which have a certain distribution (complementary or identical); a morpheme is a sequence of one or more phonemes with a certain distribution; a word is a sequence of one or more morphemes with a certain distribution; a phrase is a sequence of one or more words with a certain distribution.
  7. This is not exactly the same notion of a phrase we know today as
  1. phrases could be discontinuous, e.g. run the battery down
  1. meaning was not supposed to be considered (because it couldn’t be observed)
  2. Phrases could be endocentric (replaceable by a word contained in the phrase) or exocentric (not replaceable by a word contained in it)

a)  he [ate the fish] he [ate]

b)  he went [to London] *he went [to] *he went [London]

  1. Generative Grammar (Chomsky)
  1. Chomsky heavily criticised Structuralism
  1. Operationalism doesn’t work in the mental sphere– too restrictive

c)  he argued that language is knowledge, not behaviour

  1. Empiricism is wrong – rationalism is the way to go
  2. using only distribution as a means of discovering linguistic facts is too restrictive – ignores a lot of possible data (intuitions)
  3. meaning must be considered when hypothesising about syntax – though they are not the same thing and facts from one cannot be transferred directly to the other.
  1. He modelled the Structuralist view as a Phrase Structure Grammar, making use of Re-write rules
  1. these weren’t exactly what Structuralists had in mind as they didn’t allow discontinuous constituents – something which Chomsky then turned into a criticism of Phrase Structure Grammar: it couldn’t handle real linguistic phenomena
  1. this turned into a whole new branch of linguistics Mathematical Linguistics, based on the question of what kinds of languages could be generated by what kinds of rules, e.g. context sensitive vs. context free rules.
  1. he argued that Phrase Structure Grammar was only part of the linguistic system, but other kinds of rules (transformations) were also needed
  2. Developments in the 1960s
  3. an early development was to separate the Lexicon from the phrase structure component of the grammar
  1. this allowed phrase structure grammar to be context free
  2. it also allowed the grammar to be much simpler
  1. Remarks on Nominalisation
  2. Harris’s raised numbers
  1. in the 1940s Harris (Chomsky’s teacher and a follower of Bloomfield) introduced a system to account for non-repeatable substitutions
  1. On distributional grounds, if X can substitute for Y, then X = Y
  2. moreover, if X+Z can substitute for Y, then X+Z = Y
  3. but in this case it should be that X+Z = X and X+Z should be able to substitute for X
  1. However, this does not always work
  1. N+hood = N because boyhood can replace life in sentences like

a)  his life was spent on an island in the Mediterranean

  1. however, we cannot replace the N in the N+hood sequence with N+hood as we never get N+hood+hood
  1. to cope with this Harris suggested that we add raised numbers of a different kind for those elements which cannot be substituted repeatedly
  1. boy = N1
    N1+hood = N2
  2. then, N2 can replace N1 in all cases except for the equation that defines N2 due to a restriction that says if Nn contains Nm then n > m
  1. for repeatable substitutions, the same raised number is given to both elements
  1. man = N1
    old man = N1
  2. because ‘man’ in ‘old man’ can be replaced by another A+N sequence (old grumpy man)
  3. On Nominalisation
  1. arguing that not all syntax is transformational
  1. issue: nominalisations were seen to be syntactically formed from underlying sentences:

a)  the enemy destroyed the city à the enemy’s destruction of the city

b)  John shoots elephants à John’s shooting elephants

  1. argument: while gerunds are productive and regular (indicating a syntactic analysis) derived nominals are neither. Therefore certain processes (derivational nominalisation) take place in the Lexicon
  2. If all the following are lexically formed, then we need phrase structure rules to generate their structures:

a)  destruction of the city (NP à N (of) NP)

b)  his reply to the interviewer (NP à N PP)

c)  their belief that he left (NP à N S)

  1. But this increases the number of rules needed as (derived) nouns will have the same number of complements that verbs have (adjectives derived from verbs will go to exasperate the problem)
  2. So a first step was to introduce the following kinds of rules:

a)  VP à V Comp

b)  NP à N Comp

c)  AP à A Comp

d)  Comp à NP/PP/S/...

  1. There are two problems with this

a)  if these rules are motivated only for derived forms, it casts doubt on the proposal as a syntactic approach could get by with the VP rules and changing the VPs into NPs and APs

b)  the rules introduce Comp as a category. But this category plays no significant part in any other syntactic rule and so it has no independent justification

  1. Problem a) is addressed by the observation that there are non-derived nouns (and adjectives) which require these structures

a)  N (of) NP (professor of chemistry)
N PP (pride in his work)
N S (fact that he left)

  1. Problem b) can be addressed if we introduce a more general notation where complements are claimed to be phrases of categories which are determined by the words they are complements of – the phrase structure rules only have to stipulate that they are phrases

a)  VP à V YP
NP à N YP
AP à A YP

  1. But once we have introduced category variables, we can make the whole system simpler:

a)  XP à X YP

  1. the next step was motivated by the fact that in derived nominals, the subject of the clause is realised as the ‘determiner’ of the NP:
  1. the enemy destroyed the city
  1. the enemy’s destruction of the city
  1. The subject is separated from the verb inside the VP in the structure of S

S
NP VP
V ...

  1. Therefore we need a structure like this for the NP

NP
Det ??
N ...

  1. Chomsky called the phrase which joins with ?? to make the whole phrase the Specifier of ??
  2. He systematised the whole thing by making use of something like Harris’s ‘raised number’ notation, but using ‘bars’:
  1. Χ à Spec Χ
    Χ à X ...
  1. This produces a structure
  2. Χ
    Spec Χ
    X ...
  1. 1970s developments
  2. Inclusion of Prepositions
  1. prepositions were excluded from the ‘on Nominalisations’ because there are no derived prepositions. They were included into the system about 3 yrs later.
  2. Head
  1. prior to X-bar there was no way in PSG to capture the structuralist notion of endocentric phrases (phrases which could be replaced by one of their elements)
  1. XP à Y Z

a)  there is no necessary connection between the properties of XP and any element that it contains, so why any Y or Z can replace XP is entirely mysterious

  1. it became apparent that the head (X) in the X-bar scheme had a special property
  1. X is necessary as without it we don’t know what X’ or X’’ are
  2. Not every X is accompanied by a complement or a specifier
  3. therefore X can be the only element in XP
  4. so X can replace XP
  1. another problem solved by the notion of a head is that arbitrary phrase structure rules predict non-occurring phrases as possible:
  1. VP à N PP

a)  the question of why VPs (or other phrases) must contain a verb (or relevant other category) is solved under X-bar theory.

  1. how many bars?
  1. Chomsky’s original proposal had phrases with 2 bars simply because this was all that was needed to do the job
  1. the job was to make NPs have a similar structure to sentences

a)  sentences have subjects and verbal complements

b)  so NPs needed specifiers and nominal complements

  1. however, there was no other reason to believe that this was all that was needed. In the years that followed, various suggestions we followed based around two questions

a)  how many bars are necessary?

b)  should all phrases have the same number of bars?

  1. X-bar theory does not predict any particular number of bars and so any limit that is imposed has to be stipulated. Empirical based stipulations are circular. In principle, a phrase could have any number of bars and all phrases could be restricted to the maximum needed to account for one type of phrase – it is just assumed that the extra bars have no purpose in most phrases. Alternatively we assume that all phrases of a particular kind have the number of bars necessary for accounting for the most complex example of that kind and so all phrases may have different numbers of bars

a)  e.g. perhaps PP has only one bar (there is no specifier of PP), AP and VP have two bars and NP has three bars.

  1. adjuncts
  1. the original theory did not mention adjuncts at all and pre-X-bar theory didn’t really have much to say on the subject either (adjuncts were simply included inside the phrase along with complements). But the special status of complements given in X-bar theory meant that something needed to be said.
  2. Jackendoff (1977)
  1. he recognised two kinds of adjunct, central and peripheral

a)  adverbs of time place and manner are central

b)  speaker oriented adverbs (sadly, worryingly, etc.)

  1. These were included at different levels to reflect the fact that peripheral adjuncts are further from the head
  1. took the recursive aspect of adjuncts to be due to the unlimited number of branches that could come from one node. This predicts that adjuncts are not hierarchically organised.
  1. Hornstein and Lightfoot
  1. pointed out that adjuncts are hierarchically organised

a)  semantically: white computer table

b)  syntactically: a man in the shop with a beard and one with a moustache – a man in the shop with a beard and one in the street with a moustache – this man in the shop with a beard and that one

c)  they claimed that a new structure was needed for adjunction which allowed recursion: N’ à N’ adjunct

d)  they supported this by pointing out that each new N’ adds a property:

i)  a student = one property (of being a student)

ii)  a student of linguistics = still one property

iii)  a student of linguistics with long hair = two properties (being a student and having long hair)

iv)  a student of linguistics with long hair who I met = three properties (being a student, having long hair and having been met by me)

  1. However, not everyone was convinced and some claimed that this kind of adjunction (adjunction in the base) should not be allowed
  2. the head of S
  1. in the original work Chomsky stated that sentences are introduced by a non-X-bar rule and so X-bar notions such as head, specifier and complement do not apply
  2. Jackendoff claimed that sentences are really VPs – headed by the verb. In this way he made VP structure the same as NP structure with a uniform three bar limit
  3. however, others at the same time started talking about the inflection being the ‘head’ of the clause – but at the time this was not really brought into X-bar theory
  4. S and S-bar
  1. in the early 1970 Bresnan argued that complementisers were not included inside S, but form a constituent with S
  1. S can be pronominalised, coordinated, gapped and even moved separately from the complementiser
  1. she called the constituent made up of the complementiser and S ‘S-bar’ = however, this was not meant to indicate that this was part of X-bar theory (S cannot be a head of anything as it is not a word)
  2. some took this literally however, and started to propose an S’’ to introduce a position for the topic – but this was misguided
  3. the status of X-bar rules
  1. In the original paper, it was not entirely clear what the status of X-bar rules should be
  2. it is clear that X-bar rules cannot account for cross categorial differences – e.g. V and P take bare NP complements, N and A cannot: V, N and A take sentential complements, P cannot.
  3. Jackendoff claimed that such observations necessitated specific Phrase structure rules (N’ à N PP ...). X-bar rules were to be seen as a template for allowable phrase structure rules.
  4. However, this was to be challenged in the early 1980s
  1. 1980s developments
  2. The status of X-bar theory
  1. Jackendoff
  1. X-bar theory as a constraint on phrase structure grammar

a)  there are cross categorial differences

i)  N and A don’t take bare NP complements

ii)  V and P do

iii)  P cannot take clausal complements

iv)  N, V, A do

v)  N takes D specifiers, A takes degree specifiers, V has subject as specifier

b)  these cannot be lexical (how could we explain why all lexical elements of a certain category behave in the same way),

c)  they can’t be to do with X-bar either (this doesn’t refer to categories)

d)  the only place which gives generalisations about categories is phrase structure rules.

  1. Stowell
  1. X-bar theory as phrase structure grammar

a)  if cross categorial differences can be explained by other principles, we can reduce the role of phrase structure rules