In HPSG, how is the dependency between 'which book' and the gap after 'read' in 'Which book did she read?' maintained without moving the wh-word from the gap site?
AA transformation moves the wh-word to sentence-initial position at the level of logical form, as in Minimalism
BThe grammar generates two independent copies of the wh-phrase and checks their agreement at the end of the derivation
CA SLASH feature propagates up the tree from the gap site, using structure-sharing so the gap and the filler refer to the same object in the feature structure
DA deletion rule removes the gap and leaves the fronted wh-word bearing all properties of the missing element
HPSG is explicitly non-derivational — there is no movement, no transformation, no derivation at all. Instead, a SLASH feature is introduced at the gap site and propagated up the tree through structure-sharing until it is discharged at the point where the filler appears. Because structure-sharing means the gap and filler refer to the same data object (not copies that must be kept in sync), their phonological, syntactic, and semantic properties are automatically identical. This is the formal payoff of the unification framework.
Question 2 Multiple Choice
A transitive verb 'devours' starts with a non-empty COMPS list. After combining with its noun phrase object, what is the state of the resulting verb phrase's COMPS list?
AUnchanged — the verb phrase inherits all feature values from the head verb, including its COMPS list
BThe object's features are appended to the COMPS list of the verb phrase
CThe object is consumed off the COMPS list, leaving the verb phrase with an empty COMPS list
DThe COMPS list is replaced by an SPR list, which will be satisfied by the subject
In HPSG, valence lists track which arguments a head still needs to combine with. When a complement combines with the head, the Valence Principle ensures the complement is 'consumed' — removed from the COMPS list of the resulting phrase. A well-formed sentence is one where all valence requirements have been discharged: every COMPS and SPR list is empty at the root. This is how HPSG encodes subcategorization without movement or phrase structure rules.
Question 3 True / False
In HPSG, the Head Feature Principle ensures that certain features of the head word are literally shared with the mother node — not merely copied to it.
TTrue
FFalse
Answer: True
Structure-sharing in HPSG means that head features and mother features refer to the same location in the data structure, not independent slots that happen to hold the same value. A change to the head's feature is therefore automatically reflected in the mother's feature — they are the same object. This is what makes HPSG a constraint-based rather than rule-based system: constraints apply simultaneously to a single feature structure, not sequentially to a derivation.
Question 4 True / False
HPSG uses movement transformations, like Minimalism's Internal Merge, to handle long-distance dependencies such as wh-questions and relative clauses.
TTrue
FFalse
Answer: False
HPSG is explicitly constraint-based and non-derivational — there are no movement operations, no derivational history, and no levels of representation that transformations operate between. Long-distance dependencies are handled by the SLASH mechanism: a feature introduced at the gap site propagates up the tree through structure-sharing until it is bound off by the filler. The grammar states constraints that a feature structure must simultaneously satisfy; it does not build structures through ordered operations.
Question 5 Short Answer
What is structure-sharing in HPSG, and why is it more powerful than simply copying feature values between parts of a structure?
Think about your answer, then reveal below.
Model answer: Structure-sharing means that two positions in a feature structure refer to the same data object — they are co-indexed to literally the same value, not independent copies that happen to match. Copying creates two separate slots that must be kept in sync by explicit checking mechanisms; structure-sharing makes identity automatic. If the shared value changes for any reason, the change is immediately reflected everywhere the value is shared. For long-distance dependencies, structure-sharing ensures that a gap and its filler have identical properties not because the grammar checks them against each other, but because they are the same thing in the data structure. This is why HPSG can handle agreement, head-feature inheritance, and wh-dependencies without stipulating separate enforcement mechanisms — identity constraints arise naturally from the architecture.
This distinction is the formal reason HPSG's unification-based architecture is particularly elegant for linguistic analysis: properties that in other frameworks require explicit rules or operations to maintain emerge automatically from the data structure itself.