We effortlessly understand sentences that are incomplete, ambiguous, or arranged in unexpected ways. My research investigates the hidden grammatical representations that make this possible. Using evidence from English and other languages, I develop formal models of syntax that explain how speakers recover omitted information, interpret unusual sentence structures, and construct meaning from remarkably little linguistic input. My broader goal is to better understand the computational principles that underlie the human language faculty.
We understand the meaning of what someone says by considering what they could have said, but didn’t. For instance, if we hear somebody utter sentence 1, we understand them to mean that they have exactly two kids, even though 1 would strictly speaking be true even if the speaker has, say, three kids.
1. I have two kids.
What we are doing (unconsciously) is negating every stronger alternative utterance of the form “I have n kids”, since, if they had n > 2 kids, they would have said so.
Ido and I ask whether syntactic well-formedness restricts this kind of pragmatic reasoning over alternative utterances. We argue that it does, based on contrasts between the interpretation of progressive -ing verbs in sentences like 2b versus those in 3b.
Notice that the sentence in 2a sounds fine while the one in 2b is odd. Sentence 2b seems to suggest that NOLA tends to move, contrary to our knowledge of the world. In terms of alternatives, the speaker could have replaced “is lying” in 2b with “lies” to get 2a; the fact that they did not indicates that they do not believe the generic/habitual meaning of the bare present form “lies” to hold.
2a. New Orleans lies at the mouth of the Mississippi River.
2b. #New Orleans is lying at the mouth of the Mississippi River.
The pair in 3 is different: it has an “inverted” word-order, in which the predicate comes first and the subject comes last. It is an independent syntactic fact that inversion requires a form of “be”. Sentence 3a is then ungrammatical, but 3b is grammatical, because it has a form of “be” in the third-person present-tense “is”. Interestingly, 3b does not sound odd in the same way that 2b does, even though all of the words are the same. Sentence 3b could be uttered as the introduction to a documentary about NOLA, for instance.
3a. *Lies at the mouth of the Mississippi River New Orleans.
3b. Lying at the mouth of the Mississippi River is (the great city of) New Orleans.
Ido and I suggest that the lack of oddness in 3b (as opposed to 2b) comes from the ungrammaticality of the alternative 3a. The listener hearing 3b could conceivably compare it to the meaning corresponding to the ungrammatical sentence in 3a. The fact that the listener does not do so indicates that syntactic well-formedness restricts the availability of utterances for pragmatic reasoning.
This gets at deep questions about the nature of pragmatics. What kinds of objects are alternatives? Are they utterances (a form of action)? Or are they syntactic objects? Or perhaps logical representations? Or might they even be concepts? If we are correct, then alternatives are syntactic.
Everyday conversation is full of expressions that leave most of a sentence unspoken. For example, if someone asks the sentence in example 1, another person can naturally reply with 2.
1. Where should we put the pie?
2. On the table.
≈ We should put the pie on the table.
Even though the words “we should put the pie” are never spoken in 2, everyone understands them. A prominent hypothesis holds that such replies involve unpronounced syntactic structure, which allows us to recover the meaning of the elliptical expression. My research investigates what these missing pieces tell us about the hidden structure of language.
My project on what I call fragment subjects investigates responses like 3. At first glance, this sentence is puzzling: a location such as the table cannot literally be “a good idea.” Instead, speakers naturally understand 3 as having roughly the same meaning as 4. This suggests that 3 also contains unspoken grammatical structure.
3. On the table would be a good idea.
4. For us to put the pie on the table would be a good idea.
Although this first explanation seems plausible, it turns out to make incorrect predictions. For example, clausal subjects can be pronounced at the right edge of a sentence, as in 6, where the subject position is replaced by expletive “it”. This same maneuver does not work with fragment subjects, as in 5.
5. *Would it be a good idea on the table?
6. Would it be a good idea for us to put the pie on the table?
I argue that fragment expressions are not all built in the same way. Some omit an entire sentence, while others omit only part of one. By studying subtle patterns in English and Greek, I show that speakers mentally reconstruct different amounts of hidden grammatical structure depending on the kind of fragment they hear.
The subject of a transitive verb in the active voice, as in 1a, typically denotes the doer or “Agent” of the activity described by the verb. When a verb is made passive, another phrase becomes the subject. In 1b we see that the direct object denoting the thing placed “moves” to the subject position. Interestingly, a prepositional phrase (PP) may become the subject, as shown in 1c; this latter construction is called “Locative Inversion”.
1a. Mark placed a bowl of pasta on the table.
1b. A bowl of pasta was placed on the table.
1c. On the table was placed a bowl of pasta.
The apparent optionality in what becomes the subject when the verb passive raises several deep questions about syntax. For instance, is syntactic movement deterministic? That is, if more than one possible mover exists, is there a rule or procedure that determines which gets to move?
Contrasts like that in 2 suggest that the position of the object is critical in allowing the PP to move. We see in 2a that the object may ordinarily precede preposition-like verb “particles” like down. Interestingly, when Locative Inversion occurs, the object must follow the particle—compare 2c with 2c’.
2a. Mark put a bowl of pasta down on the table.
2b. A bowl of pasta was put down on the table.
2c. *On the table was put a bowl of pasta down.
2c'. On the table was put down a bowl of pasta.
My research on locative inversion investigates the relationship between syntactic structure, linear order, and locality. A central question is whether the structural height of different potential movers determines which phrase ends up moving, or whether it is simply the leftmost eligible phrase that moves. In Feldcamp (2026), I argue that the latter hypothesis is closer to the truth. The crucial data comes from locative inversion configurations in which the structurally higher of two potential movers is to the right of the structurally lower of the two.
The paradigm in 3 is parallel to that in 2, but the PP starts out as a modifier of the verb “prove”, higher than the object noun phrase in the embedded clause. Nonetheless, the NP must be pronounced at the right edge of the sentence, if PP is to move. What we find is that the phrase on the right cannot move unless the phrase on the left first moves to the right of it. I develop and argue for a theory of locality that explains this generalization.
3a. Professor Zhang proved the Twin Prime Conjecture false in the lecture.
3b. The Twin Prime Conjecture was proved false in the lecture.
3c. *In the lecture was proved the Twin Prime Conjecture false.
3c. In the lecture was proved false—the world-famous Twin Prime Conjecture, which has baffled generations of mathematicians.
An earlier version of this paper was presented as a poster at WCCFL 42.
Work on noun phrase syntax across languages has often focused on the role of nominal linkers—functional elements that appear to license phrases as modifiers, arguments, or other dependents of a head noun. The Japanese morpheme -na, for instance, obligatorily follows a certain class of adjectives when they are used as adnominal modifiers (1a), but does not occur otherwise (1b).
1a. kirei *(na) ie (Japanese)
clean LNK house
'a clean house'
2a. kono ie wa kirei (*na) desu
this house TOP clean LNK be
'This house is clean.'
This paper is an initial attempt at developing an explanatory theory of syntactically optional, or “weak” dependency, viewed primarily through the lens of its functional structure in linker constructions. The central idea is that modification and other weak dependent constructions involve the same structure-building operations through which arguments are integrated into the clause, namely the application of set Merge and feature-checking local to a selecting head.
Bayesian models have become a powerful tool for understanding how children learn language, but applying these methods to modern syntactic theories has remained challenging. My research develops a mathematical framework that assigns probabilities to the syntactic trees generated by Minimalist Grammars, a widely studied formalization of Chomsky’s Minimalist Program. I show that these probabilities can be normalized into a genuine probability distribution, providing a foundation for Bayesian inference over minimalist grammars. This work helps connect contemporary syntactic theory with probabilistic models of language acquisition and grammar learning.
For the technically-inclined, a handout is attached. I introduce a new bottom-up prior over Minimalist Grammar derivations and prove that it is normalizable. The proof uses bounds on the number of unary-binary tree shapes derived from the theory of Motzkin numbers. To my knowledge, this combinatorial method has not previously been applied to Minimalist Grammars. The resulting probability distribution provides a basis for Bayesian inference over competing grammars.
This work-in-progress aims to provide a missing mathematical foundation for probabilistic models of modern syntactic theory, making it possible to compare competing grammars within a principled Bayesian framework.