An official website of the United States government
Recent work in subregular syntax has revealed deep parallels among syntactic phenomena, many of which fall under the computational class TSL (Graf, 2018, 2022). Vu et al. (2019) argue that case dependencies are yet another member of this class. But their analysis focuses mainly on English, which is famously case-poor. In this paper I present a TSL analysis of Japanese, which features a much wider range of case-marking patterns, adding support to the claim that case dependencies, and by extension syntactic dependencies, are TSL.
more »
« less
This content will become publicly available on January 30, 2026
Tier-based strict locality and the typology of agreement
This paper presents a subregular analysis of syntactic agreement patterns modeled using command strings over Minimalist Grammar (MG) dependency trees (Graf and Shafiei 2019), incorporating a novel MG treatment of agreement. Phenomena of interest include relativized minimality and its exceptions, direction of feature transmission, and configurations involving chains of agreeing elements. Such patterns are shown to fall within the class of tier-based strictly 2-local (TSL-2) languages, which has previously been argued to subsume the majority of long-distance syntactic phenomena, as well as those in phonology and morphology (Graf 2022a). This characterization places a tight upper bound on the range of configurations that are predicted to occur while providing parameters for variation which closely match the observed typology.
more »
« less
- Award ID(s):
- 1845344
- PAR ID:
- 10608497
- Publisher / Repository:
- Institute of Computer Science Polish Academy of Sciences Warsaw
- Date Published:
- Journal Name:
- Journal of Language Modelling
- Volume:
- 13
- Issue:
- 1
- ISSN:
- 2299-856X
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Many long-distance linguistic dependencies across domains can be modeled as tier-based strictly local (TSL) patterns (Graf 2022a). Such patterns are in principle efficiently learnable, but known algorithms require unrealistic conditions. Heuser et al. (2024) present an algorithm for learning syntactic islands which involves memorizing local bigrams along attested paths; no tiers are involved. I propose a method for inferring tier membership which augments the latter algorithm to produce a TSL grammar, and show that this model also derives a version of the Height-Locality Connection (Keine 2019).more » « less
-
Lexical selection, functional hierarchies, and adjunct ordering are arguably distinct parts of syntax, yet are surprisingly similar in their computational properties. All three fall within the formal class strictly local (SL) and thus are maximally simple. Many phonological patterns are also SL, motivating a more detailed comparison. Towards this end, I develop a model based on com- mand strings (Graf & Shafiei 2019) which allows syntactic and phonological grammars to be visualized using finite-state automata. Using this model, I show that the same basic patterns allowed within SL occur in both domains.more » « less
-
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-tot-tree transduction, defined here for the first time. The restriction to ISL transductions correctly rules out various deviant category systems.more » « less
-
null (Ed.)We introduce The Benchmark of Linguistic Minimal Pairs (BLiMP), 1 a challenge set for evaluating the linguistic knowledge of language models (LMs) on major grammatical phenomena in English. BLiMP consists of 67 individual datasets, each containing 1,000 minimal pairs—that is, pairs of minimally different sentences that contrast in grammatical acceptability and isolate specific phenomenon in syntax, morphology, or semantics. We generate the data according to linguist-crafted grammar templates, and human aggregate agreement with the labels is 96.4%. We evaluate n-gram, LSTM, and Transformer (GPT-2 and Transformer-XL) LMs by observing whether they assign a higher probability to the acceptable sentence in each minimal pair. We find that state-of-the-art models identify morphological contrasts related to agreement reliably, but they struggle with some subtle semantic and syntactic phenomena, such as negative polarity items and extraction islands.more » « less
