Phrase-Structure Rules inside Transformational–Generative Grammar

Review of Previous Session and Course Road-Map

• The previous (6th) lecture dealt with Immediate-Constituent (IC) analysis, originated by Leonard Bloomfield (1930s) and later refined with Zellig Harris.
• Weaknesses of IC noted:
  • No principled way to mark complements vs. adjuncts; all constituents appear on the same tier.
  • Discontinuities, coordination, ambiguous or naturally occurring sentences often cannot be handled.
  • Requires drawing a separate tree for every single sentence; poor descriptive and computational economy.
• Current block (Sessions 7–8) covers Generative–Transformational Grammar (TGG):
  • Session 7 = Phrase Structure Rules (PSG).
  • Session 8 = Transformational Rules.
  • Two further sessions will treat X-Bar Theory (1980s) and post-GB developments (up to ≈ 2022).

Immediate-Constituent vs. Phrase Structure Example

Consider the three sentences:

1. The boy read the novel.
2. The boy read the novel very loudly.
3. The boy read the novel very loudly in the class.

IC Analysis:

• Break successively into constituents: (The boy) (read the novel) → The / boy & read / the / novel etc.
• „the novel", „very loudly", „in the class" are siblings → no hierarchy, no distinction between complement and adjunct.

PSG Analysis:
$S \rightarrow NP\;VP$
$VP \rightarrow V\ (NP)\ (PP)\ (ADVP)$

• „the novel" = complement (required to satisfy verb READ’s argument structure).
• „very loudly", „in the class" = adjuncts (optional; appear lower in the tree).
• The tree makes complement/adjunct status and hierarchical ordering explicit.

Generative–Transformational Grammar (TGG)

• Introduced by Noam A. Chomsky, $1957$ (book Syntactic Structures; elaborated in Aspects of the Theory of Syntax 1965).
• Central claim: humans possess an innate, rule-governed capacity to generate an infinite set of sentences from a finite set of rules (language is linked to human cognition).
• Two rule systems inside TGG:
  1. Phrase-Structure Rules (a.k.a. Rewrite or Production rules).
  2. Transformational Rules (move, delete, insert constituents, etc.).

Components of Phrase Structure Grammar (PSG)

1. Lexicon – list of words with syntactic category, sub-category, selectional properties.
2. Rewrite/Production Rules – context-free schemata such as $NP \rightarrow Det\ N$.
3. Syntactic Categories – $S, NP, VP, PP, ADJP, ADVP …$
4. Head – obligatory word that determines phrase label (N is head of NP, V of VP, etc.).
5. Nodes in a tree:
   • Mother, Daughter, Sister relations (like family tree).
   • Terminal nodes = words/lexemes.
6. Hierarchy – top node $S$, successive projections downward.

Order Convention When Writing Rules

1. Begin with $S$.
2. Expand $VP$.
3. If present, expand $PP$.
4. Expand $NP$.
5. Finally, adjunct phrases (ADJP, ADVP, etc.).

Worked Example ▸ "The boy went to the park"

Phrase-Structure Tree outline:
\begin{aligned}
S &\rightarrow NP\;VP\
NP &\rightarrow Det\;N\
VP &\rightarrow V\;PP\
PP &\rightarrow P\;NP\
NP &\rightarrow Det\;N
\end{aligned}
Lexical Rules:
$Det \rightarrow \text{the}$
$N \rightarrow \text{boy}\ \ |\ \ \text{park}$
$V \rightarrow \text{went}$
$P \rightarrow \text{to}$

Derivation Steps (each step replaces the left-most non-terminal using the numbered rule):

1. $S$
2. $NP\;VP$
3. $Det\;N\;VP$
4. $Det\;N\;V\;PP$
5. $Det\;N\;V\;P\;NP$
6. $Det\;N\;V\;P\;Det\;N$
7. Replace terminals → the boy went to the park.

Types of PSG Rules

• Categorial Rules – combine abstract categories (context-free):
$S \rightarrow NP\;VP$
$VP \rightarrow V\;NP$
• Lexical Rules – map categories onto concrete words:
$Det \rightarrow the/ a/ this$
$V_{past} \rightarrow went/ loved/ ate$

Productivity vs. Restriction

• Pattern $Det\ N\ V\ Det\ N$ can generate infinite sentences:
  The girl sang a song; The man wrote the letter; …
• Adding lexical rules restricts possibilities, but syntactic acceptability is still larger than semantic acceptability.

Sub-Categorisation & Selectional Restrictions

• Sub-categories further refine categories (e.g. $N<em>{prop}, N</em>{pron}, V<em>{trans}, V</em>{intrans}$).
• Argument Structure: verbs specify number & type of complements.
  • cry 〈NP[agent]〉 (intransitive) → “The baby cried.” complete.
  • kick 〈NP[agent], NP[theme]〉 (transitive) → “The boy kicked.” is incomplete; needs object.
• Selectional Restrictions ensure semantic fit (animacy, concreteness, thematic roles):
  • Structurally OK but semantically odd: The bone ate the dog.
  • Violates verb’s selectional rule: EAT expects an animate subject, edible object.

Advantages of PSG over IC Analysis

1. Clear hierarchical representation; complements vs. adjuncts visible.
2. General, flexible, rule-economical – one rule set generates many sentences.
3. Formal & systematic – suitable for computational models (NLP, parsers, MT).
4. Captures syntax–semantics interface via sub-categorisation & selectional rules.
5. Strong descriptive power – handles complex, coordinated, or discontinuous constructions.
6. Modular – rules reusable across sentences.
7. Pedagogical value – clarifies subject/verb/object patterns; aids error diagnosis.
8. Compatible with later theories: Government-Binding, Minimalism, Dependency, Case & Theta Theory, etc.

Key Terminology & Symbols

• $S$ – Sentence clause node.
• $NP, VP, PP, ADJP, ADVP$ – phrasal categories.
• $Det, N, V, P$ – lexical categories.
• Head – obligatory element determining phrase label.
• Complement – obligatory sister to head (licensed by sub-cat frame).
• Adjunct – optional modifier phrase.
• Argument – NP (or clause/PP) demanded by predicate’s theta grid.
• Rewrite arrow $\rightarrow$ – “is expanded as”.
• Hash # – silent boundary markers before and after sentence.
• Terminal node – lexeme; non-terminal – category symbol.

Connections & Implications

• IC analysis provided first structural insight; PSG refines this with explicit hierarchy and rule generativity.
• PSG lays groundwork for transformational component: once base structure is built, transformations can move/delete constituents to yield surface forms (e.g. Passive, Questions, WH-movement).
• Sub-categorisation & selectional mechanisms anticipate later developments such as Theta Theory, Case Theory, X-Bar schema, and Minimalist feature checking.
• Computational linguistics (e.g. context-free parsing, Treebank annotation) still largely rests on PSG notions.

Practical Study Tips

1. Practise drawing trees and writing the corresponding rule list + derivation.
2. Always state category rules first, lexical rules second.
3. Respect the conventional order: $S\rightarrow VP\rightarrow PP\rightarrow NP …$
4. Test grammaticality:
   • Syntactic test → matches rule inventory.
   • Semantic test → satisfies selectional restrictions.
5. Identify complements vs. adjuncts with deletion and obligatoriness tests.
6. Memorise common sub-categorisation frames (e.g. V[ditrans] 〈NP, NP〉 = give).