The recent collection of articles, The Art of Translation [Masterstvo perevoda], follows publication of two other books on the subject, The Problems of Translation of Creative Writing [Voprosy khudozhestvennogo perevoda] and another work also bearing the title The Art of Translation.

The Theory and Practice of Translation

Finite State Morphology

Greek-English Lexicon of the New Testament Based on Semantic Domains

Despite a long and venerable tradition in descriptive linguistics, dependency grammar has until recently played a fairly marginal role both in theoretical linguistics and in natural language processing. The increasing interest in dependency-based representations in natural language parsing in recent years appears to be motivated both by the potential usefulness of bilexical relations in disambiguation and by the gains in efficiency that result from the more constrained parsing problem for these representations. In this paper, we will review the state of the art in dependency-based parsing, starting with the theoretical foundations of dependency grammar and moving on to consider both grammar-driven and data-driven methods for dependency parsing. We will limit our attention to systems for dependency parsing in a narrow sense, i.e. systems where the analysis assigned to an input sentence takes the form of a dependency structure. This means that we will not discuss systems that exploit dependency relations for the construction of another type of representation, such as the head-driven parsing models of Collins (1997, 1999). Moreover, we will restrict ourselves to systems for full parsing, which means that we will not deal with systems that produce a partial or underspecified representation of dependency structure, such as Constraint Grammar parsers (Karlsson, 1990; Karlsson et al., 1995).

/pdf/dependency-grammar-and-dependency-parsing-el5mqhox9a.pdf

Dependency Grammar and Dependency Parsing

We propose a hypothesis only baseline for diagnosing Natural Language Inference (NLI). Especially when an NLI dataset assumes inference is occurring based purely on the relationship between a context and a hypothesis, it follows that assessing entailment relations while ignoring the provided context is a degenerate solution. Yet, through experiments on ten distinct NLI datasets, we find that this approach, which we refer to as a hypothesis-only model, is able to significantly outperform a majority class baseline across a number of NLI datasets. Our analysis suggests that statistical irregularities may allow a model to perform NLI in some datasets beyond what should be achievable without access to the context.

Hypothesis Only Baselines in Natural Language Inference

Inquiries into words, constraints and contexts

Sentence-level representations are necessary for various natural language processing tasks. Recurrent neural networks have proven to be very effective in learning distributed representations and can be trained efficiently on natural language inference tasks. We build on top of one such model and propose a hierarchy of bidirectional LSTM and max pooling layers that implements an iterative refinement strategy and yields state of the art results on the SciTail dataset as well as strong results for Stanford Natural Language Inference and Multi-Genre Natural Language Inference. We can show that the sentence embeddings learned in this way can be utilized in a wide variety of transfer learning tasks, outperforming InferSent on 7 out of 10 and SkipThought on 8 out of 9 SentEval sentence embedding evaluation tasks. Furthermore, our model beats the InferSent model in 8 out of 10 recently published SentEval probing tasks designed to evaluate sentence embeddings’ ability to capture some of the important linguistic properties of sentences.

/pdf/sentence-embeddings-in-nli-with-iterative-refinement-3nrz6l9epf.pdf

Sentence Embeddings in NLI with Iterative Refinement Encoders

The paper discusses a language operation that we call context restriction. This operation is closely associated with context restriction rules (Koskenniemi, 1983; Kiraz, 2000), right-arrow rules or implication rules (Koskenniemi et al., 1992; Voutilainen, 1997) and the restriction operator (Beesley and Karttunen, 2003). The operation has been used in finite-state phonology and morphology in certain limited ways. A more general setting involves restricting overlapping occurrences of a center language under context conditions. Recently, star-free regular languages (and all regular languages) have been shown to be closed under context restrictions with such “overlapping centers” (Yli-Jyra, 2003), but the construction involved is overly complex and becomes impractical when the number of operands grows. In this paper, we improve this recent result by presenting a more practical construction. This construction is not only simpler but it also leads to a generalization where contexts and centers may appear as conditions at both sides of the implication arrow (⇒): licensing conditions on the right-hand side specify the restriction and triggering conditions on the left-hand side regulate activation of the restriction. One application of the generalization is to facilitate splitting certain context restriction rules in grammars into a conjunction of separate rules.

/pdf/compiling-contextual-restrictions-on-strings-into-finite-1yj4x1mwag.pdf

Compiling contextual restrictions on strings into finite-state automata

The number of treebanks available for different languages is growing steadily. A considerable portion of the recent treebanks use annotation schemes that are based on dependency syntax. In this paper, we give a model for linguistically adequate classes of dependency structures in treebanks. Our model is tested using the Danish Dependency Treebank [13]. The modern dependency syntax was pioneered by Tesniere [27]. His core concepts, binary dependencies and unique heads, are mostly shared in the recent dependency syntactic theories [6, 9, 18, 24, 26]. These theories stress the functional structure, while paying much less attention to linear word order. Lecerf’s projectivity hypothesis [15, 17] assumes a constraint on linear wordorder in dependency analyses. It says that if a word A depends directly on a word B and some word C intervenes between them in linear order, then C depends directly on A or on B or some other intervening word[23]. The projectivity constraint has been a popular simplification in many computational dependency grammars since 1960’s [4]. It does not only make parsing algorithms simple and efficient [16], but also equips us with neat ways to visualize analyses, with an equivalent, constituent based representation for dependency trees [4, 7], and with a criterion for stylistically marked analyses [27, 22] and for abnormal information structure [10]. The tendency that sentences admit projective analyses has been observed in many languages, including French [15], Swedish [20], Finnish [1], and Turkish [21]. Unfortunately, projectivity does not lend itself to adequate treatment of certain non-local syntactic phenomena which are extensively studied in the literature of constituent-based theories such as TG, GB, GPSG, TAG, and LFG. Among these

/pdf/multiplanarity-a-model-for-dependency-structures-in-45av7ou1r2.pdf

Multiplanarity - a model for dependency structures in treebanks

Recurrent neural networks have proven to be very effective for natural language inference tasks. We build on top of one such model, namely BiLSTM with max pooling, and show that adding a hierarchy of BiLSTM and max pooling layers yields state of the art results for the SNLI sentence encoding-based models and the SciTail dataset, as well as provides strong results for the MultiNLI dataset. We also show that our sentence embeddings can be utilized in a wide variety of transfer learning tasks, outperforming InferSent on 7 out of 10 and SkipThought on 8 out of 9 SentEval sentence embedding evaluation tasks. Furthermore, our model beats the InferSent model in 8 out of 10 recently published SentEval probing tasks designed to evaluate sentence embeddings' ability to capture some of the important linguistic properties of sentences.

Anssi Yli-Jyrä

Papers

Inquiries into words, constraints and contexts

Sentence Embeddings in NLI with Iterative Refinement Encoders

Compiling contextual restrictions on strings into finite-state automata

Multiplanarity - a model for dependency structures in treebanks

Natural Language Inference with Hierarchical BiLSTM Max Pooling Architecture.