Recent work has shown that language models (LMs) trained on synthetic corpora can exhibit typological preferences that resemble crosslinguistic regularities in human languages, particularly for syntactic phenomena such as word order.In this paper, we extend this paradigm to differential argument marking (DAM), a semantic licensing system in which morphological marking depends on semantic prominence.Using a controlled synthetic learning method, we train GPT-2 models on 18 corpora implementing distinct DAM systems and evaluate their generalization using minimal pairs.Our results reveal a dissociation between two typological dimensions of DAM.Models reliably exhibit human-like preferences for natural markedness direction, favoring systems in which overt marking targets semantically atypical arguments.In contrast, models do not reproduce the strong object preference in human languages, in which overt marking in DAM more often targets objects rather than subjects.These findings suggest that different typological tendencies may arise from distinct underlying sources. 1

Abstract This paper introduces Filtered Corpus Training, a method that trains language models (LMs) on corpora with certain linguistic constructions filtered out from the training data, and uses it to measure the ability of LMs to perform linguistic generalization on the basis of indirect evidence. We apply the method to both LSTM and Transformer LMs (of roughly comparable size), developing filtered corpora that target a wide range of linguistic phenomena. Our results show that while transformers are better qua LMs (as measured by perplexity), both models perform equally and surprisingly well on linguistic generalization measures, suggesting that they are capable of generalizing from indirect evidence.

This work presents methods for learning crosslingual sentence representations using paired or unpaired bilingual texts.We hypothesize that the cross-lingual alignment strategy is transferable, and therefore a model trained to align only two languages can encode multilingually more aligned representations.We thus introduce dual-pivot transfer: training on one language pair and evaluating on other pairs.To study this theory, we design unsupervised models trained on unpaired sentences and single-pair supervised models trained on bitexts, both based on the unsupervised language model XLM-R with its parameters frozen.The experiments evaluate the models as universal sentence encoders on the task of unsupervised bitext mining on two datasets, where the unsupervised model reaches the state of the art of unsupervised retrieval, and the alternative single-pair supervised model approaches the performance of multilingually supervised models.The results suggest that bilingual training techniques as proposed can be applied to get sentence representations with multilingual alignment.

We propose a general framework to study language emergence through signaling games with neural agents.Using a continuous latent space, we are able to (i) train using backpropagation, (ii) show that discrete messages nonetheless naturally emerge.We explore whether categorical perception effects follow and show that the messages are not compositional.

The ADC method of proof search in propositional natural deduction proposed in 2000 proceeds bottom up first by Analysing the sequent into sub-goals by applying all possible introduction rules, and then by checking whether each of these sub-goals can be established using only elimination rules ( Direct Chaining ). This looks simpler than the worst case complexity (PSPACE) of the derivability problem for intuitionistic propositional logic. We investigate the complexity of ADC for various fragments. ADC derivability is polynomially decidable for the &, →-fragment by a generalization of a familiar direct chaining algorithm for Horn formulas. Adding ∨ leads to a CoNP-complete fragment. A short counterexample in case the goal sequent is not ADC derivable is provided by witnessing all relevant disjunctions. Adding constant ⊥ preserves CoNP completeness.

A method for automated location of shape differences in diseased anatomical structures via high resolution biomedical atlases annotated with labels from formal ontologies is described. In particular, a high resolution magnetic resonance image of the myocardium of the human left ventricle was segmented and annotated with structural terms from an extracted subset of the Foundational Model of Anatomy ontology. The atlas was registered to the end systole template of a previous study of left ventricular remodeling in cardiomyopathy using a diffeomorphic registration algorithm. The previous study used thresholding and visual inspection to locate a region of statistical significance which distinguished patients with ischemic cardiomyopathy from those with nonischemic cardiomyopathy. Using semantic technologies and the deformed annotated atlas, this location was more precisely found. Although this study used only a cardiac atlas, it provides a proof-of-concept that ontologically labeled biomedical atlases of any anatomical structure can be used to automate location-based inferences.