Questions: Feature Geometry and Underspecification

Question 1

English nasal assimilation changes 'in-' to 'im-' before bilabials (impossible) and to 'iŋ-' before velars (incomplete). Why does the entire place specification change rather than just one feature?

• A. Each language independently stipulates which features must assimilate together
• B. All place features are dominated by a single Place node, so assimilation applies to the whole node and moves all dominated features simultaneously
• C. The nasals /m/, /n/, and /ŋ/ share the same underlying representation and only surface differently
• D. Assimilation rules operate on segments as units, not on individual features, so all features change

Answer: (B)

In feature geometry, [labial], [coronal], and [dorsal] features are all dominated by the Place node. When place assimilation occurs, it targets the Place node — the nasal's Place specification is replaced by that of the following consonant, automatically including all features beneath it. This is why assimilation is holistic: the geometry encodes the dependency. A flat-matrix approach would need special stipulations; feature geometry makes holistic behavior a structural prediction.

Question 2

In a language where all obstruents are voiced by default and only a few are underlyingly voiceless, what does underspecification predict about how voicing is stored in the mental lexicon?

• A. Every voiced obstruent is marked [+voiced] and every voiceless obstruent is marked [−voiced] in the lexicon
• B. Only the voiceless obstruents are specified as [−voiced]; voiced obstruents are unvalued for voicing and receive [+voiced] by a default rule
• C. Neither voiced nor voiceless obstruents are specified for voicing; surface values are entirely computed by context
• D. The lexicon marks all obstruents as [+voiced] since that is the majority pattern

Answer: (B)

Underspecification exploits predictability: when a feature value can be assigned by a rule (voiced is default), it need not be stored. The lexicon marks only the non-default, unpredictable value — [−voiced] for voiceless exceptions. Voiced obstruents are left unvalued; the default rule assigns [+voiced] to all segments lacking a voicing specification. This reduces memorization without losing information. Option D would mean voiced obstruents ARE specified, defeating the economy of underspecification.

Question 3

A segment that has no Place specification in its underlying representation is better positioned to undergo place assimilation than a fully specified segment.

• T. True
• F. False

Answer: True

A segment with no underlying Place value has nothing to preserve or conflict with, so it can freely receive the Place node of its neighbor through assimilation. A fully specified segment would create a structural conflict — two Place values competing — which is typically resolved by blocking assimilation or causing deletion. Underspecification thus directly predicts which segments will be 'chameleon-like': those with absent specifications are the most transparent to spreading processes.

Question 4

Underspecification proposes that some phonological features are absent from the grammar largely — they do not exist for sounds that appear to lack them.

• T. True
• F. False

Answer: False

This is a key misconception. Underspecification does not claim features are absent from the phonological system — it claims they are not valued in underlying representations, but remain as structural positions that can be filled by rules, defaults, or assimilation. An 'absent' feature could never be filled in; an 'unspecified' feature is a structural slot waiting to receive a value. Feature geometry and underspecification model how these slots get filled predictably.

Question 5

How does hierarchical feature organization explain why phonological rules systematically affect natural classes of sounds rather than arbitrary groupings?

Model answer: In a flat feature matrix, any subset of features could in principle be targeted by a rule, including arbitrary combinations. Feature geometry organizes features into hierarchical nodes corresponding to articulatory groupings — the Place node, the Laryngeal node, the Manner node. A phonological rule can only target a node (a natural class defined by shared hierarchical position) or a feature beneath it. This structural constraint makes it impossible to write a rule targeting an arbitrary mix of features from different nodes. The hierarchy encodes the phonetic knowledge that sounds behave as natural classes — bilabials, velars, nasals — rather than arbitrary feature bundles.

The empirical prediction is that phonological processes respect articulatory organization: you find rules spreading all place features together (targeting the Place node) but not rules spreading [labial] plus [nasal] while leaving other place features alone. Feature geometry converts this observation from an unexplained coincidence into a structural prediction.