Thomas Graf

Curbing Feature Coding: Strictly Local Feature Assignment

2020-01-04T00:00:00-05:00

Graf, Thomas

Abstract Graf (2017) warns that every syntactic formalism faces a severe overgeneration problem because of the hidden power of subcategorization. Any constraint definable in monadic second-order logic can be compiled into the category system so that it is indirectly enforced as part of subcategorization. Not only does this kind of feature coding deprive syntactic proposals of their empirical bite, it also undermines computational efforts to limit syntactic formalisms via subregular complexity. This paper presents a subregular solution to feature coding. Instead of features being a cheap resource that comes for free, features must be assigned by a transduction. In particular, category features must be assigned by an input strictly local (ISL) tree-to-tree transduction, defined here for the first time. The restriction to ISL transductions correctly rules out various deviant category systems.

Files [pdf]

:::bibtex
@misc{Graf20SCiLtalk,
    author = {Graf, Thomas},
    title = {Curbing Feature Coding: Strictly Local Feature Assignment},
    year = {2020},
    note = {Slides of a talk given at \emph{{SCiL} 2020}, January 2--5, co-located with the {LSA 2020}}
}

A Subregular Bound on the Complexity of Lexical Quantifiers

2019-12-18T00:00:00-05:00

Graf, Thomas

Abstract Semantic automata theory studies the complexity of generalized quantifiers in terms of the string languages that describe their truth conditions. An important point has gone unnoticed so far: for most quantifiers that are determiners, these string languages are subregular. Whereas quantifier phrases such as an even number of correspond to specific regular languages, every, no, some and not all stay within the much weaker class of tier-based strictly local languages (TSL). In addition, it seems to hold crosslinguistically that a TSL quantifier may be expressed by a single lexical item only if its tier projection is monotonic. This constitutes a novel typological universal that limits how complex a meaning may be stored in a single lexical item. TSL is also known to play a central role in phonology, morphology, and syntax, which suggests that subregularity in general and tier-based strict locality in particular may be a computational universal of natural language that surfaces across all its submodules, including semantics.

Files [pdf]

:::bibtex
@misc{Graf19ACtalk,
    author={Graf, Thomas},
    title={A Subregular Bound on the Complexity of Lexical Quantifiers},
    year = {2019},
    note = {Slides of a talk given at the Amsterdam Colloquium, {D}ecember 18--20, University of Amsterdam, Amsterdam, Netherlands},
}

A Subregular Bound on the Complexity of Lexical Quantifiers

2019-12-15T00:00:00-05:00

Graf, Thomas

Files [pdf]

:::bibtex
@inproceedings{Graf19AC,
    author = {Graf, Thomas},
    title = {A Subregular Bound on the Complexity of Lexical Quantifiers},
    year = {2019},
    booktitle = {Proceedings of the 22nd {A}msterdam {C}olloquium},
    editor = {Schl\"{o}der, Julian J. and McHugh, Dean and Roelofsen, Floris},
    pages = {455--464},
}

Subregular Linguistics for Linguists

2019-11-08T00:00:00-05:00

Graf, Thomas

Abstract Drawing from computational work that is known as the subregular program, I argue against two received views in linguistics: “phonology and syntax are very different’ and “subcategorization is a solved problem”.

Cognitive parallelism
Subregular notions of complexity can be applied to strings as well as trees. Doing so reveals that phonology and syntax are remarkably similar (and those parallels even extend into morphology and semantics). For instance, islands and blocking effects are instances of the same computational mechanism.
Subcategorization
Subcategorization (or c-selection) is rarely studied by linguists, but it is actually a source of tremendous overgeneration. Once again subregular notions of complexity can be used to address this problem. This isn’t just a mathematical exercise, but makes concrete empirical predictions about the nature of category systems, subcategorization, the status of empty heads, the DP-analysis, DM-style roots, and once again highlights parallels to phonology.

The general upshot is that subregular concepts, despite their computational origin, are intuitive and linguistically fertile: they address conceptual issues, bridge gaps between linguistic subfields, and make concrete empirical predictions. Subregular linguistics is just linguistics with some computational flavor sprinkled on top.

Disclaimer: This talk is 100% formula-free.

Files [pdf]

:::bibtex
@misc{Graf19UMasstalk,
    author = {Graf, Thomas},
    title = {Subregular Linguistics for Linguists},
    year = {2019},
    note = {Invited talk, November 8, University of Massachusetts Amherst, Amherst, Massachusetts}
}

Features: More Trouble Than They’re Worth?

2019-05-27T00:00:00-04:00

Graf, Thomas

Abstract Do contemporary syntactic theories rely on too many features? Decades of computational research have culminated in two contradictory answers:

Yes, there are too many features because having at least two features is already one too many and brings in undesirable overgeneration.
No, the number of features is irrelevant because a grammar with no features can be just as powerful as one with thousands of features.

The source of this confusing state of affairs is the intimate relation between features and constraints. Features can be replaced by constraints, to a point where the grammar is completely feature-free. And constraints can be compiled down into the feature system until no constraints are left and everything is done by feature checking.

Proposition 1: Features and constraints are interchangeable.

The power to express arbitrary constraints causes massive overgeneration, which must be curtailed. There is a substantive body of computational work on how constraints can be limited in a linguistically natural fashion, whereas it is largely unclear how the same can be achieved with restrictions on feature systems. I will sketch what options have been explored so far and why constraints are easier to rein in than features.

Proposition 2: The power of constraints is easy to restrict, that of features is not.

In the absence of meaningful restrictions on features, then, constraint-based accounts should be preferred. I present small case studies on successive-cyclic movement, selection, and morphosyntax that illustrate what this approach looks like in practice and how it can yield new theoretical and empirical insights.

Files [pdf]

:::bibtex
@misc{Graf19Tromsotalk,
    author = {Graf, Thomas},
    title = {Features: More Trouble Than They're Worth?},
    year = {2019},
    note = {Invited talk, May 27, Workshop on \emph{Thirty Million Theories of Syntactic Features}, University of Troms\o, Troms\o, Norway}
}

Diagnosing Movement via the Absence of c-Command Relations

2019-03-23T00:00:00-04:00

Graf, Thomas, and Sabine Laszakovits

Abstract Based on an earlier finding that almost all c-command dependencies amount to subregular constraints on c-strings, we argue that all exceptions to this generalization involve movement. Since regulating movement is already known to be beyond the purview of c-strings, it is only natural that phenomena that deeply interact with movement may require additional mechanisms beyond c-strings.

Files

:::bibtex
@misc{GrafLaszakovits19PLCtalk,
    author = {Graf, Thomas and Laszakovits, Sabine},
    title = {Diagnosing Movement via the Absence of c-Command Relations},
    year = {2019},
    note = {Slides presented at the 43rd Penn Linguistics Colloquium (PLC43), March 22--24,
            University of Pennsylvania, Philadelphia, PA}
}

C-Command Dependencies as TSL String Constraints

2019-01-05T00:00:00-05:00

Graf, Thomas, and Nazila Shafiei

Abstract We provide a general formal framework for analyzing c-command based dependencies in syntax, e.g. binding and NPI licensing, from a subregular perspective. C-command relations are represented as strings computed from Minimalist derivation trees, and syntactic dependencies are shown to be input-output tier-based strictly local over such strings. The complexity of many syntactic phenomena thus is comparable to dependencies in phonology and morphology.

Files [pdf]

:::bibtex
@misc{GrafShafiei19SCiLposter,
    author = {Graf, Thomas and Shafiei, Nazila},
    title = {C-Command Dependencies as {TSL} String Constraints},
    year = {2019},
    note = {Poster presented at \emph{{SCiL} 2019}, January 3--6, co-located with the {LSA 2019}}
}

The Computational Cost of Generalizations: An Example from Micromorphology

2019-01-05T00:00:00-05:00

Moradi, Sedigheh, Alëna Aksënova, and Thomas Graf

Abstract The central concern of linguistics is to succinctly state generalizations. But as numerous linguists have pointed out over the years, generalizations do not always come for free. A formalism’s ability to account for the data does not entail that it can express the relevant generalizations about the data. We illustrate this point with a computational case study in micro-morphology. More precisely, we show that the subregular complexity of adjective inflection in Noon varies depending on what generalizations one wishes to capture.

Files [pdf]

:::bibtex
@misc{MoradiEtAl19SCiLposter,
    author = {Moradi, Sedigheh and Aks\"{e}nova, Al\"{e}na and Graf, Thomas},
    title = {The Computational Cost of Generalizations: An Example from Micromorphology},
    year = {2019},
    note = {Poster presented at \emph{{SCiL} 2019}, January 3--6, co-located with the {LSA 2019}}
}