Sentences are represented concretely as a linear string of words.
However we parse them in a hierarchical tree structure, with each of the nodes being constituents.
Experimental evidence: Merrill Garrett’s click experiments.
- Clicks are placed in a neutral position in the sentence
- Participants tended to perceive the clicks at constituent boundaries instead of the actual word’s position.
Constituents are defined by rules that dictate what sort of components they can comprise of.
They are written in the forms:
\[\begin{align} XP &\rightarrow X \ Y \ Z \\ XP &\rightarrow (Opt) \ X \\ XP &\rightarrow (MinOnce+) \ X \\ \end{align}\]The head of a phrase gives its phrase its category.
In $XP \rightarrow X \ Y \ Z$, $X$ is the head of the phrase $XP$.
Informal formulation: Modifiers are always attached to the phrase they modify.
Formal formulation: If an XP modifies some head Y, then XP must be a sister to Y (i.e., a daughter of YP).
Structure of the sentence or tense phrase.
\[TP \rightarrow NP \ (T) \ VP\]Clauses can be embedded. This requires a complementizer
\[CP \rightarrow (C) \ TP\]Can the head of a phrase be optional??
CPs can also modify N.
The man [whose car I hit __ last week] sued me.
Gap is the “extraction site” from which the NP “var” after the wh-word is the filler.
\[RP \rightarrow (R) \ NP \ (NP) \ VP\]Observe that rules are allowed to be recursive.
This accounts partially for the infinite nature of language.
It is an active area of research to find languages not demonstration recursion.
Purpose: We should find instances where groups of words behave as single units.
Tests are such instances (where word groups can be swapped out for substitutes).
Tests may not always be conclusive, or they may fail. This is because of certain quirks of the individual tests.
e.g.
The subject island: constituents cannot be moved out of the subject island.
[Bruce loved] and [Kelly hated] phonology class.
“Bruce loved” and “Kelly hated” are not constituents.
This is because “phonology class” is elided from “Bruce loved”.
e.g.
Zheng put the bagel [in the freezer]. It was [in the freezer] that Zheng put the bagel. Zheng put there bagel [there]. Where did Zheng put the bagel? [In the freezer].
Zheng put [the bagel in the freezer]. *It was [the bagel in the freezer] that Zheng put. *Zheng put [it]. *[The bagel in the freezer] was what Zheng put.
Constituents of different categories behave differently.
Left-headed languages tend to have $X$ appear to the left within $XP$s, right-headed languages tend to have them appear to the right.
There are also polysynthetic and free-word order languages.
Sentences represented as a hierarchical order of units are geometric objects.
If syntactic trees are geometric objects, they can be studied and described mathematically – the focus of this chapter.
Parent*
relation
Parent*OfAll
relation. A exhaustively dominates a set of nodes U iff U is a set of terminal nodes and A dominates every $U \in A$.
Parent relation
Branches cannot cross.
the phenomenon of Binding.
I.e. the following configs are allowed in govermment:
Grammatical relations are determined by language specific structural rules.
In English:
This view proposes that syntax comes first and semantics are what we interpreted of it.