A grammar specifies that to form a grammatical sentence, the subject and verb feature structures must unify. The subject is specified as [NUM: singular, PERS: 3rd] and the verb requires [NUM: plural, PERS: 3rd, TENSE: past]. What is the result?
AUnification succeeds because person features match, and the NUM conflict is resolved in favor of the verb
BUnification fails because the NUM values conflict, correctly predicting the sentence is ungrammatical
CUnification partially succeeds, yielding a combined structure that marks the sentence as pragmatically unusual
DUnification succeeds because the verb's specification is more complete and overrides the subject
Unification requires that ALL shared attributes be compatible. The NUM attribute has conflicting values (singular vs. plural) — this is a contradiction that cannot be resolved. Unification fails, and the grammar correctly predicts the sentence is ungrammatical. This is precisely the elegance of the mechanism: there is no special rule for number agreement, no procedure checking subject against verb — the general unification operation handles it automatically, and failure equals ungrammaticality.
Question 2 Multiple Choice
What is the primary advantage of modeling agreement through unification rather than writing separate procedural rules for each agreement type (e.g., 'if subject is singular, make verb singular')?'
AUnification is faster to compute and requires less memory than rule-based approaches
BUnification applies only to syntactic features, so semantic interpretation can proceed independently
CA single declarative unification constraint handles all agreement phenomena uniformly — the grammar specifies what must match, and unification applies the logic without special-casing each agreement type
DUnification eliminates the need for feature structures by encoding agreement in word order
The power of unification is that it is both declarative (stating what must be true) and general (applicable to any combination of feature structures). Instead of writing separate rules for number agreement, gender agreement, case agreement, person agreement, etc. — each with their own procedural logic — you specify features once and let unification do the checking. Adding a new agreement dimension (say, animacy in a language that distinguishes animate/inanimate agreement) just means adding a new feature attribute; the unification mechanism itself does not change.
Question 3 True / False
Unification can succeed even when two feature structures have conflicting values for a shared attribute, provided the conflict is on a minor or optional feature.
TTrue
FFalse
Answer: False
Unification is all-or-nothing: it succeeds if and only if ALL shared attributes are compatible. There is no concept of 'minor' or 'optional' conflicts in the formal operation — any conflict on any shared attribute causes unification to fail. This strictness is what gives the mechanism its predictive power: grammaticality is a binary outcome (unification succeeds or fails), not a matter of degree. If a grammar needs to allow some flexibility, it must do so by leaving certain features underspecified (variables), not by tolerating conflicts.
Question 4 True / False
An underspecified feature — one represented as a variable that has not yet been assigned a specific value — can unify with any specific value, allowing grammatical specifications to be completed through combination with other elements.
TTrue
FFalse
Answer: True
Underspecification is the mechanism that allows unification to handle partial information. A feature structure representing an English verb that can appear with either singular or plural subjects might leave NUM underspecified. When this verb combines with a singular subject [NUM: singular], unification fills in the variable with 'singular,' yielding a fully specified combined structure. This is how agreement propagates through a sentence: underspecified elements receive values from elements they combine with, and conflicts emerge only when two structures try to assign different specific values to the same attribute.
Question 5 Short Answer
How does unification provide a declarative account of grammaticality? Explain why a unification failure corresponds to an ungrammatical sentence.
Think about your answer, then reveal below.
Model answer: In a unification-based grammar, every grammatical combination is licensed by a successful unification of the feature structures of the combining elements. The grammar does not contain explicit lists of grammatical sentences or procedural rules that generate them — it contains feature constraints, and unification checks whether those constraints are mutually satisfiable. When a combination violates a constraint (e.g., subject and verb disagree in number), the feature structures conflict and unification fails. The grammar predicts the sentence is ungrammatical as a logical consequence of this failure, not because a rule explicitly bans it. Grammaticality is thus reduced to a question of constraint satisfaction.
The declarative/procedural distinction is central here. A procedural grammar says: 'follow these steps to generate a sentence.' A declarative grammar says: 'a sentence is grammatical if and only if these constraints are satisfied.' Unification implements the checking half of the declarative approach. The result is a grammar that is both formally explicit (every prediction follows from the feature specifications) and easy to extend (new phenomena add new features, not new procedures).