Yuri Balashov

The success of deep learning in natural language processing raises intriguing questions about the nature of linguistic meaning and ways in which it can be processed by natural and artificial systems. One such question has to do with subword segmentation algorithms widely employed in language modeling, machine translation, and other tasks since 2016. These algorithms often cut words into semantically opaque pieces, such as ‘period’, ‘on’, ‘t’, and ‘ist’ in ‘period|on|t|ist’. The system then represents the resulting segments in a dense vector space, which is expected to model grammatical relations among them. This representation may in turn be used to map ‘period|on|t|ist’ (English) to ‘par|od|ont|iste’ (French). Thus, instead of being modeled at the lexical level, translation is reformulated more generally as the task of learning the best bilingual mapping between the sequences of subword segments of two languages; and sometimes even between pure character sequences: ‘p|e|r|i|o|d|o|n|t|i|s|t’ → ‘p|a|r|o|d|o|n|t|i|s|t|e’. Such segmentations and alignments are at work in highly efficient end-to-end machine translation systems, despite their allegedly opaque nature. But do they have linguistic or philosophical plausibility? I attempt to cast light on this question, in the spirit of making artificial intelligence more transparent and explainable.

I offer an argument in defense of four-dimensionalism, the view that objects are temporally, as well as spatially extended. The argument is of the inference-to-the-best-explanation variety and is based on relativistic considerations. It deals with the situation in which one and the same object has different three-dimensional shapes at the same time and proceeds by asking what sort of thing it must be in order to present itself in such different ways in various “perspectives” (associated with moving reference frames) without being different from itself. I argue that the best answer is that the object must be four-dimensional. It will then have differing 3D shapes in different perspectives because such shapes are intrinsic properties of its 3D parts.

Neural machine translation (NMT) is one of the success stories of deep learning and artificial intelligence. Revolutionary innovations in the computational architectures made in 2015–2017 have led to dramatic improvements in the quality of machine translation (MT) and changed the field forever. Some professional translators welcome these changes with enthusiasm, others less so. But everyone has to deal with them. Historically, the relationship between human translation and MT has been uneasy and complicated, but an increasing number of players in both fields are now coming to view it as synergistic. OPUS-CAT is a collection of machine translation software based on a Windows implementation of Marian NMT—a stable production-ready neural machine translation framework used in many industrial and research applications. It can be installed on your PC, fine-tuned to specialized translation domains, and integrated into CAT tools.

The success of deep learning in natural language processing raises intriguing questions about the nature of linguistic meaning and ways in which it can be processed by natural and artificial systems. One such question has to do with subword segmentation algorithms widely employed in language modeling, machine translation, and other tasks since 2016. These algorithms often cut words into semantically opaque pieces, such as ‘period’, ‘on’, ‘t’, and ‘ist’ in ‘period|on|t|ist’. The system then represents the resulting segments in a dense vector space, which is expected to model grammatical relations among them. This representation may in turn be used to map ‘period|on|t|ist’ (English) to ‘par|od|ont|iste’ (French). Thus, instead of being modeled at the lexical level, translation is reformulated more generally as the task of learning the best bilingual mapping between the sequences of subword segments of two languages; and sometimes even between pure character sequences: ‘p|e|r|i|o|d|o|n|t|i|s|t’ → ‘p|a|r|o|d|o|n|t|i|s|t|e’. Such subword segmentations and alignments are at work in highly efficient end-to-end machine translation systems, despite their allegedly opaque nature. The computational value of such processes is unquestionable. But do they have any linguistic or philosophical plausibility? I attempt to cast light on this question by reviewing the relevant details of the subword segmentation algorithms and by relating them to important philosophical and linguistic debates, in the spirit of making artificial intelligence more transparent and explainable.

I examine the issue of persistence over time in the context of the special theory of relativity (SR). The four-dimensional ontology of perduring objects is clearly favored by SR. But it is a different question if and to what extent this ontology is required, and the rival endurantist ontology ruled out, by this theory. In addressing this question, I take the essential idea of endurantism, that objects are wholly present at single moments of time, and argue that it commits one to unacceptable conclusions regarding coexistence, in the context of SR. I then propose and discuss a plausible account of coexistence for perduring objects, which is free of these defects. This leaves the endurantist room for some maneuvers. I consider them and show that they do not really help the endurantist out. She can accommodate the notion of coexistence in the relativistic framework only at the cost of renouncing central endurantist intuitions.

Debates about material coincidence tend to start with common-sense intuitions but quickly leave them behind and lead to highly problematic conclusions. Reconciling the latter with common sense is the next stage in the process, which often requires revision of some of the initial beliefs and has been used to adjudicate many rather abstract and technical proposals in the metaphysics of composition and persistence, ranging from natural (constitutionalism) to radical (nihilism). I have no disagreement with this overall strategy: theories do need to turn abstract at some point, move beyond common sense, and eventually force upon us interesting, novel, and often counterintuitive revisions in our overall conceptual scheme. This applies to all theoretical areas, and contemporary metaphysics is no exception. But while the latter is widely regarded as being quite extreme in this respect, I want to argue that, in one sense, it is not extreme enough. I do it by developing a new case of material coincidence that is initially motivated not by common sense but by physical considerations, and is not susceptible to any of the standard solutions. One lesson of science is its ability to expand our imagination beyond the limits of common sense. This may have importance for metaphysics too.

Cosmology as Weltanschauung is as old as the world. Cosmology as a physical discipline, however, is a child of this century, born in 1917, when Albert Einstein and Willem de Sitter first applied the theory of general relativity to the space-time of the entire universe. When did the child come of age and become a fully-fledged science? A popular myth shared by many practitioners holds that this did not happen until 1965, when the discovery of the 2.7K cosmic microwave background radiation validated the concept of the hot big bang. Earlier accomplishments culminating in the 1965 event include the derivation of evolving relativistic models by Alexander Friedmann and George Lemaître, Edwin Hubble’s discovery of the redshift-magnitude relationship and George Gamow’s theory of the primordial nucleosynthesis containing the prediction of the microwave background left from the hot and dense past of the universe. This linear development is thrown into sharper relief in textbooks, by referring to an inarticulate..