AI summary by TwinMind
Lecture 12: Syntax, Part 2
MIT OpenCourseWare
1h 10m 49s
Syntax Tree Construction Review
- Started with morphology example "unlockable" - three morphemes composed in pairs
- Two ways to construct: [un + lock] + able vs. [lock + able] + un with different meanings
- Applied same pairwise merge operation concept to syntax
- Sentences are strings of words composed via pairwise merge operation
- Example sentence: "I will find the book in the garage" (8 words)
Tree Building and Labeling Conventions
- Merge determiners "the" with nouns "book" and "garage" to form larger objects
- Labeling rules for nodes created by merge:
- In morphology: affixes provide instructions (un + verb → verb, able + verb → adjective)
- In syntax: label of result is always label of one of the two merged items
- Label reflects what syntactic rules care about (nouns can be modified various ways)
- Phrase notation conventions:
- P (phrase) = highest node with given label from specific word
- Unmarked = lowest thing with label (just the word)
- Everything between = bar notation (V-bar, N-bar, etc.)
- Bar notation originated from typewriter era with actual bars over letters
- Never called "V Prime" - always "V bar"
Prepositional Phrase Attachment and Ambiguity
- Prepositional phrases can modify both noun phrases and verb phrases
- Creates recursive structures: NPs can contain PPs, PPs can contain NPs
- Allows arbitrarily long phrases (competence vs. performance distinction)
- Classic ambiguity example: "I shot an elephant in my pajamas"
- Two interpretations: elephant wearing pajamas vs. speaker wearing pajamas
- Two different tree structures based on PP attachment point
Constituency Tests - Topicalization
- Topicalization: moving constituents to sentence beginning for emphasis
- Test reveals which structures are possible:
- "The elephant I shot in my pajamas" - only speaker wearing pajamas reading
- "The elephant in my pajamas I shot" - only elephant wearing pajamas reading
- Topicalization collapses ambiguity by forcing specific tree structure
- Only constituents can be topicalized
- Additional constraint: may require phrases (not just bars) for topicalization
Multiple Prepositional Phrases
- Example: "I shot the elephant in my pajamas in a tuxedo"
- Only one interpretation possible due to structural constraints
- Elephant wearing pajamas, speaker wearing tuxedo
- No line-crossing allowed in trees (Ghostbusters rule)
Complete Sentence Structure
- Added tense phrase (TP) for complete sentences
- "Will" labeled as T (tense) - covers auxiliaries like "can," "might"
- Different from adverbs like "quickly" based on:
- Question formation: "Will the girl find it?" vs. "*Quickly the girl found it?"
- Ellipsis patterns: "The girl will" vs. "*The girl quickly"
- Sentence structure: [The girl [will [find the book in the garage]]]
- Merge T with VP first, then merge subject with result
VP Ellipsis Phenomenon
- Process of silencing verb phrases in certain contexts
- Examples: "Who will find the book? The girl will [VP]"
- Distinguished from constituency structure
- Allows omission of repeated VP content
Additional Ambiguity Examples
- "I will tickle the child with the feather"
- Same PP attachment ambiguity pattern
- Child has feather vs. using feather to tickle
- Topicalization tests confirm different structures
Tree Terminology
- Sister: two nodes merged together to create larger node
- Mother: node created by merging two nodes
- Daughter: nodes that were merged to create mother node
- Immediate domination: direct mother-daughter relationship
- Domination: transitive closure of immediate domination
- Constituent: words dominated by single node
Pronoun Reference Constraints (Preview)
- "She thinks I like Mary" - she cannot refer to Mary
- "Mary thinks I like her" - her can refer to Mary
- "Everyone who knows her likes Mary" - her can refer to Mary
- Not simply about linear order in sentence
- Structural relationship determines pronoun reference possibilities
- Will provide additional constituency tests
