Linguistic Theory

1
Linguistic Theory

• Lecture 2
• Phrase Structure

2
What was there before structure?

• Classical studies
• Languages such as Latin
• Rich morphology
• Not many word order restrictions
• Mainly associated with meaning
• Words carried meaning
• Therefore main emphasis on word form paradigms
  and meaning, but not much on syntax

3

• Traditional Grammar
• Prescriptive claimed Latin to be pure language
  and all others were deviations
• Application of what was known about classical
  languages to modern languages
• Thus, phenomena that were not similar to Latin
  (word order restrictions) were ignored or seen as
  deviant

4

• 1800s
• As early as the late 1500s similarities between
  Indian and European languages had been noted
• But the idea of an Indo-European family of
  languages did not become popular until the start
  of the 1800s. This was a new era for
  linguistics
• Again the concentration was on classical
  languages (Greek, Latin, Sanskrit)
• Most easily accessible data concerned word forms
  (phonology/morphology)
• Once more, syntax was not well studied

5

• In most of these approaches, the main aspects of
  syntactic description concerned grammatical
  functions (subject, object, etc.)
• In the languages being studied, grammatical
  functions were mainly indicated morphologically
  (Case, agreement, etc.)
• Grammatical functions were defined semantically
• E.g. Subject the one who performs the
  action/what the sentence is about
• Therefore grammatical functions were associated
  with words and no phrases were necessary

6

• Subordination of clauses was recognised
• So the basis of constituent structure was
  available (one thing containing other things)
• But the need to extend this to non-clausal groups
  of words (phrases) didnt seem to arise

7
Structuralism

• Empiricist view discovery procedures
  constituent structure analysis
• Positive points
• Could account for distributional patterns
• John/poor John left the room
• I saw John/poor John
• Could account for the FACT of distribution

8

• Negative points
• Was not formalised or made specific in terms of
  rules which constituted a grammar
• Such a grammar would be in the mind, which didnt
  exist
• Therefore did not consider the issue of
  restricting grammatical theory to make it more
  explanatory
• Why do phrases distribute the way they do?

9

• Linguistic Relativity
• Popular view of the structuralists
• Each grammar could only be studied relative to
  itself i.e. No commonalities between languages
  to compare them
• Supported by observation of Amerindian languages
• Therefore, in principle, languages could be
  anyhow
• They were, more or less, an accident of the
  environment and culture they were set in

10

• We determine what is a phrase in a language on
  distributional grounds
• But we cannot say why certain groups of words are
  phrases because in principle any group of words
  could be a phrase
• Description is easy
• The English gerund is an NP because it
  distributes like an NP
• I dont like careful plans/carefully making
  plans
• careful plans are/carefully making plans is
  difficult
• But the gerund, unlike most NPs, contains no
  obvious noun
• NP Adv carefully V making NP plans
• Explanation is impossible
• Why do most NPs contain a noun?
• Why is in the park not an NP?

11

• Why is a noun phrase a noun phrase?
• Partly on distributional grounds
• NPs distribute the same as SOME nouns
• But pronouns distribute like determiners and NPs
  distribute like pronouns so NPs distribute like
  SOME determiners!!!
• From tradition
• On the basis of semantic salience, traditional
  grammars took nouns to be grammatical functions
• Even though structuralists eschewed meaning, it
  seems that they still took the classical position
  to heart

12

• So why is a PP a PP?
• The preposition is hardly the most semantically
  salient part of a preposition phrase
• Distribution shows that they are not NPs
• What distinguishes them from NPs is that they
  contain a preposition, therefore we call them
  preposition phrases
• This is not very consistent
• But it doesnt matter because there is no need
  for consistency as, in principle, anything is
  possible

13
Chomsky and Phrase Structure Grammar

• Change to rationalism and the study of language
  as knowledge
• We can only study internal language by hypothesis
  testing
• Therefore we need grammatical hypotheses (
  grammar) to be explicit
• Requires formalising rules

14

• Phrase Structure Rules
• E.g. VP ? V NP
• A list of such rules Phrase Structure Grammar
• A PSG produces a set of Phrase Markers (tree
  diagrams)
• Still nothing wrong with
• NP ? Adv V NP
• So still unconstrained not explanatory
• But ...

15

• Formalising PS rules helped to
• Point out problems with basic Immediate
  Constituent Analysis to show that more is needed
  to account for human languages
• Further our understanding of rule systems and to
  suggest what is needed to capture natural
  language data

16

• Problems for IC analysis E.g. The passive
• We might analyse a passive sentence thus
• S NP VPJohn Verb
  Aux V was liked
• This is easy to do with PS rules
• But...

17

• This analysis fails to capture certain facts
• Only transitive verbs can appear in the passive
• He was liked
• he was smiled
• Transitive verbs are usually restricted to VPs
  which contain objects
• They liked John
• they liked
• Intransitives are restricted to VPs without an
  object
• He smiled
• he smiled them
• So in the passive, things are turned upside down
• Why?

18

• Verbs place semantic restrictions on their
  subjects and objects
• John hates insincerity
• insincerity hates John
• Insincerity worries John
• John worries insincerity
• Hate ? animate/sentient subject
• Worry ? animate/sentient object
• In the passive these restrictions are reversed
• Insincerity was hated (by John)
• John was hated (by insincerity)
• John was worried (by insincerity)
• insincerity was worried (by John)
• A phrase structure analysis cannot account for
  these observations

19

• Mathematical Linguistics
• Formalising rules for phrase structure using
  re-write rules enabled Chomsky to explore aspects
  of grammar that had never been thought of before
• Rule a) produces structure B)
• a) X ? Y Z B) X Y Z
• But what does rule b) produce?
• b) W X ? Y Z
• Such rules produce structures which cannot be
  represented by a tree diagram

20

• Phrase Structure Grammar rules can have only one
  element to the left of the arrow
• Unrestricted Re-write System rules can have more
  than one element to the left of the arrow
• Note that a PSG is a (restricted) kind of URS
• So, any language that a PSG can generate, a URS
  can too. But not vice versa
• URS languages PSG languages

21

• There are grammars which are more restrictive
  than URSs but less restrictive than simple PSGs
• A X ? A Y Z
• This rule rewrites X, when it is preceded by A
• The rule is context sensitive
• A context sensitive rule is a kind of
  unrestricted rewrite rule where only one element
  on the left of the arrow is rewritten
• A context free rule is a kind of context
  sensitive rule with no context stated
• So unrestricted rewrite grammar
  context sensitive PS grammar context
  free PS grammar

22

• The question is what kind of grammar is human
  grammar?
• This could be a way of restricting linguistic
  hypotheses and making them more explanatory
• Most linguists think that the phrase structure
  part of human grammar is no more complex than a
  context free rewrite system
• Though all agree that this is not enough to
  account entirely for all grammatical phenomena

23
More restrictions

• Phrase structure grammars (even context sensitive
  ones) still allow things that we dont find in
  human languages
• PP ? Adv V
• Phrases have heads
• There is no way to represent the notion of the
  head in a standard rewrite rule because what is
  on the left of the arrow is not connected to what
  is on the right
• X ? Y Z

24

• In 1970 Chomsky proposed a restriction on rewrite
  rules which addressed this problem
• X-bar theory
• Xn ? ... Xm ... (ignore n and m)
• Rules are restricted to the type where there must
  be an element to the right of the arrow which is
  the same category as the one on the left
• X-bar grammars do not form a subset of any of the
  types in the Chomsky hierarchy
• So perhaps mathematical types are not so useful
  afterall!

25
Generalised Phrase Structure Grammar

• Some argued that Chomsky was wrong about the
  limitations of context free PSGs
• The problem Chomsky pointed out was that elements
  often appear in one position when we would expect
  them to be in another (e.g. Passive)

26

• But we can overcome this problem if we allow a
  new kind of category
• X/Y
• Called a slash category
• Means an X which lacks a Y
• S NP S/NPwho NP
  VP/NP John V NP/NP
  met ?

27

• Rules needed
• X/X ? ? basic rule
• Y/X ? ... Z/X .. inheritance rule
• Y ? X Z/X resolution rule