The ability to consolidate information of different types is at the core of intelligence, and has tremendous practical value in allowing learning for one task to benefit from generalizations learned for others.In this paper we tackle the challenging task of improving semantic parsing performance, taking UCCA parsing as a test case, and AMR, SDP and Universal Dependencies (UD) parsing as auxiliary tasks.We experiment on three languages, using a uniform transition-based system and learning architecture for all parsing tasks.Despite notable conceptual, formal and domain differences, we show that multitask learning significantly improves UCCA parsing in both in-domain and out-of-domain settings.Our code is publicly available.

Sentence splitting is a major simplification operator.Here we present a simple and efficient splitting algorithm based on an automatic semantic parser.After splitting, the text is amenable for further fine-tuned simplification operations.In particular, we show that neural Machine Translation can be effectively used in this situation.Previous application of Machine Translation for simplification suffers from a considerable disadvantage in that they are overconservative, often failing to modify the source in any way.Splitting based on semantic parsing, as proposed here, alleviates this issue.Extensive automatic and human evaluation shows that the proposed method compares favorably to the stateof-the-art in combined lexical and structural simplification.

BLEU is widely considered to be an informative metric for text-to-text generation, including Text Simplification (TS).TS includes both lexical and structural aspects.In this paper we show that BLEU is not suitable for the evaluation of sentence splitting, the major structural simplification operation.We manually compiled a sentence splitting gold standard corpus containing multiple structural paraphrases, and performed a correlation analysis with human judgments.1 We find low or no correlation between BLEU and the grammaticality and meaning preservation parameters where sentence splitting is involved.Moreover, BLEU often negatively correlates with simplicity, essentially penalizing simpler sentences.

Elior Sulem, Omri Abend, Ari Rappoport. Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Papers). 2018.

Semantic representation is receiving growing attention in NLP in the past few years, and many proposals for semantic schemes (e.g., AMR, UCCA, GMB, UDS) have been put forth.Yet, little has been done to assess the achievements and the shortcomings of these new contenders, compare them with syntactic schemes, and clarify the general goals of research on semantic representation.We address these gaps by critically surveying the state of the art in the field.

We present the first parser for UCCA, a cross-linguistically applicable framework for semantic representation, which builds on extensive typological work and supports rapid annotation.UCCA poses a challenge for existing parsing techniques, as it exhibits reentrancy (resulting in DAG structures), discontinuous structures and non-terminal nodes corresponding to complex semantic units.To our knowledge, the conjunction of these formal properties is not supported by any existing parser.Our transition-based parser, which uses a novel transition set and features based on bidirectional LSTMs, has value not just for UCCA parsing: its ability to handle more general graph structures can inform the development of parsers for other semantic DAG structures, and in languages that frequently use discontinuous structures.

This paper is concerned with identifying contexts useful for training word representation models for different word classes such as adjectives (A), verbs (V), and nouns (N).We introduce a simple yet effective framework for an automatic selection of class-specific context configurations.We construct a context configuration space based on universal dependency relations between words, and efficiently search this space with an adapted beam search algorithm.In word similarity tasks for each word class, we show that our framework is both effective and efficient.Particularly, it improves the Spearman's ρ correlation with human scores on SimLex-999 over the best previously proposed class-specific contexts by 6 (A), 6 (V) and 5 (N) ρ points.With our selected context configurations, we train on only 14% (A), 26.2% (V), and 33.6% (N) of all dependency-based contexts, resulting in a reduced training time.Our results generalise: we show that the configurations our algorithm learns for one English training setup outperform previously proposed context types in another training setup for English.Moreover, basing the configuration space on universal dependencies, it is possible to transfer the learned configurations to German and Italian.We also demonstrate improved per-class results over other context types in these two languages.

The run time complexity of state-of-theart inference algorithms in graph-based dependency parsing is super-linear in the number of input words (n).Recently, pruning algorithms for these models have shown to cut a large portion of the graph edges, with minimal damage to the resulting parse trees.Solving the inference problem in run time complexity determined solely by the number of edges (m) is hence of obvious importance.We propose such an inference algorithm for first-order models, which encodes the problem as a minimum spanning tree (MST) problem in an undirected graph.This allows us to utilize state-of-the-art undirected MST algorithms whose run time is O(m) at expectation and with a very high probability.A directed parse tree is then inferred from the undirected MST and is subsequently improved with respect to the directed parsing model through local greedy updates, both steps running in O(n) time.In experiments with 18 languages, a variant of the first-order MSTParser (McDonald et al., 2005b) that employs our algorithm performs very similarly to the original parser that runs an O(n 2 ) directed MST inference.

State-of-the-art word embeddings, which are often trained on bag-of-words (BOW) contexts, provide a high quality representation of aspects of the semantics of nouns.However, their quality decreases substantially for the task of verb similarity prediction.In this paper we show that using symmetric pattern contexts (SPs, e.g., "X and Y") improves word2vec verb similarity performance by up to 15% and is also instrumental in adjective similarity prediction.The unsupervised SP contexts are even superior to a variety of dependency contexts extracted using a supervised dependency parser.Moreover, we observe that SPs and dependency coordination contexts (Coor) capture a similar type of information, and demonstrate that Coor contexts are superior to other dependency contexts including the set of all dependency contexts, although they are still inferior to SPs.Finally, there are substantially fewer SP contexts compared to alternative representations, leading to a massive reduction in training time.On an 8G words corpus and a 32 core machine, the SP model trains in 11 minutes, compared to 5 and 11 hours with BOW and all dependency contexts, respectively.

We present a novel word level vector representation based on symmetric patterns (SPs).For this aim we automatically acquire SPs (e.g., "X and Y") from a large corpus of plain text, and generate vectors where each coordinate represents the cooccurrence in SPs of the represented word with another word of the vocabulary.Our representation has three advantages over existing alternatives: First, being based on symmetric word relationships, it is highly suitable for word similarity prediction.Particularly, on the SimLex999 word similarity dataset, our model achieves a Spearman's ρ score of 0.517, compared to 0.462 of the state-of-the-art word2vec model.Interestingly, our model performs exceptionally well on verbs, outperforming stateof-the-art baselines by 20.2-41.5%.Second, pattern features can be adapted to the needs of a target NLP application.For example, we show that we can easily control whether the embeddings derived from SPs deem antonym pairs (e.g.(big,small)) as similar or dissimilar, an important distinction for tasks such as word classification and sentiment analysis.Finally, we show that a simple combination of the word similarity scores generated by our method and by word2vec results in a superior predictive power over that of each individual model, scoring as high as 0.563 in Spearman's ρ on SimLex999.This emphasizes the differences between the signals captured by each of the models.

Work on authorship attribution has traditionally focused on long texts.In this work, we tackle the question of whether the author of a very short text can be successfully identified.We use Twitter as an experimental testbed.We introduce the concept of an author's unique "signature", and show that such signatures are typical of many authors when writing very short texts.We also present a new authorship attribution feature ("flexible patterns") and demonstrate a significant improvement over our baselines.Our results show that the author of a single tweet can be identified with good accuracy in an array of flavors of the authorship attribution task.

Current social media research mainly focuses on temporal trends of the information flow and on the topology of the social graph that facilitates the propagation of information. In this paper we study the effect of the content of the idea on the information propagation. We present an efficient hybrid approach based on a linear regression for predicting the spread of an idea in a given time frame. We show that a combination of content features with temporal and topological features minimizes prediction error.

We present a novel approach for discovering word categories, sets of words sharing a significant aspect of their meaning. We utilize meta-patterns of high-frequency words and content words in order to discover pattern candidates. Symmetric patterns are then identified using graph-based measures, and word categories are created based on graph clique sets. Our method is the first pattern-based method that requires no corpus annotation or manually provided seed patterns or words. We evaluate our algorithm on very large corpora in two languages, using both human judgments and WordNet-based evaluation. Our fully unsupervised results are superior to previous work that used a POS tagged corpus, and computation time for huge corpora are orders of magnitude faster than previously reported.

In the parametric feature based design paradigm, most features possess arguments that are subsets of the boundary of the current model, subsets defined interactively by user selection of boundary entities. Any system for feature-based data exchange (FBDE) must support the exchange of such selections. In this paper we describe in detail an algorithm for supporting onedimensional selections (sets of edges and curves) for FBDE. The algorithm is applicable to a wide class of FBDE architectures, including the Universal Product Representation (UPR) and the STEP parametrics specification.

Global view-dependent illumination phenomena, in particular reflections, greatly enhance the realism of computer-generated imagery.Current interactive rendering methods do not provide satisfactory support for reflections on curved objects.In this paper we present a novel method for interactive computation of reflections on curved objects.We transform potentially reflected scene objects according to reflectors, to generate virtual objects.These are rendered by the graphics system as ordinary objects, creating a reflection image that is blended with the primary image.Virtual objects are created by tessellating scene objects and computing a virtual vertex for each resulting scene vertex.Virtual vertices are computed using a novel space subdivision, the reflection subdivision.For general polygonal mesh reflectors, we present an associated approximate acceleration scheme, the explosion map.For specific types of objects (e.g., linear extrusions of planar curves) the reflection subdivision can be reduced to a 2-D one that is utilized more accurately and efficiently.

Interactive modeling of 3-D solids is an important and difficult problem in computer graphics. The Constructive Solid Geometry (CSG) modeling scheme is highly attractive for interactive design, due to its support for hierarchical modeling and Boolean operations. Unfortunately, current algorithms for interactive display of CSG models require expensive special-purpose hardware that is not easily available. In this paper we present a method for interactive display of CSG models using standard, widely available graphics hardware. The method enables the user to interactively modify the affine transformations associated with CSG sub-objects. The application we focus upon is that of conceptual design, a stage in the design process in which rapid, interactive visualization of the model and high-level design operations are of crucial importance, while the objects are relatively simple. The method converts the CSG graph to a novel Convex Differences Aggregate(CDA) representation. The CDA utili...

The authors introduce the notion of compatible star decompositions of simple polygons. In general, given two polygons with a correspondence between their vertices, two polygonal decompositions of the two polygons are said to be compatible if there exists a one-to-one mapping between them such that the corresponding pieces are defined by corresponding vertices. For compatible star decompositions, they also require correspondence between star points of the star pieces. Compatible star decompositions have applications in computer animation and shape representation and analysis. They present two algorithms for constructing compatible star decompositions of two simple polygons. The first algorithm is optimal in the number of pieces in the decomposition, providing that such a decomposition exists without adding Steiner vertices. The second algorithm constructs compatible star decompositions with Steiner vertices, which are not minimal in the number of pieces but are asymptotically worst-case optimal in this number and in the number of added Steiner vertices. They prove that some pairs of polygons require /spl Omega/(n/sup 2/) pieces, and that the decompositions computed by the second algorithm possess no more than O(n/sup 2/) pieces. In addition to the contributions regarding compatible star decompositions, the paper also corrects an error in the only previously published polynomial algorithm for constructing a minimal star decomposition of a simple polygon, an error which might lead to a nonminimal decomposition.

Abstract Digital typefaces for computer graphics and multimedia applications should be capable of supporting operations such as font variations, transformations. deformations and blending. A powerful implementation of such operations must rely on the inherent typographic attributes of the typeface. However, even today's most advanced typeface representations support only geometric outline representations and basic font variations. In this paper we discuss high‐level typeface representations which we term Parametric Typographic Representations (PTRs). We present an algorithm for automatically extracting typographic elements of typefaces from their outline representation, which, is an essential initial step in converting typefaces from outline representations to PTRs. The extracted typographic elements include serifs, bars. sterns, slants, bows, arcs, curve stems and curve bars. Most notable is the treatment of serifs, which are represented by finite‐automata. The algorithm only needs to learn a serif type once, and is then capable of automatically recognizing it in different typefaces. We show an application of a PTR for automatic high‐quality hinting of fonts, which is one of the most important stages in, digital font production. Our system was used to generate hints for dozens of thousands of Kanji, Roman and Hebrew characters.

Abstract Designers and creative artists use computer graphics and image processing effects on stall photographs in application areas such as advertising entertainment broadcasting and the arts Most of the effects available in research arid commercial work are two‐dimensional in nature, for example image processing filters [blur, edge enhancement) and creative effects (tilings, reflections) There is almost no usage of information taken from the 3‐D world in which the objects appearing an the image are located. In this paper we present a novel method for creating 3‐D effects on photographs or in general on any image created by rendering a 3‐D world The artist interacts with the image using a set of intuitive direct manipulation interface objects These objects let the user define a 3‐D model, display at, and manipulate it in a 3‐D space which is correlated with that of the input image. The generated model can be an arbitrarily complex 3‐D polyhedron Any texture, including texture taken from the input photograph, can be mapped into any of its faces arid used for special effects We discuss and show examples for effects such as copy and paste, motion blur, model editing and deformations lighting effects, and shadows.

Some important trends in geometric modeling are the reliance on solid models rather than surface-based models and the enhancement of the expressive power of models, by using free-form objects in addition to the usual geometric primitives and by incorporating physical principles. An additional trend is the emphasis on interactive performance. In this paper, we integrate all of these requirements into a single geometric primitive by endowing the tri-variate tensor-product free-form solid with several important physical properties, including volume and internal deformation energy. Volume preservation is of benefit in several application areas of geometric modeling, including computer animation, industrial design and mechanical engineering. However, previous physics-based methods, which have usually used some form of "energy", have neglected the issue of volume (or area) preservation. We present a novel method for modeling an object composed of several tensor-product solids while preserving the desired volume of each primitive and ensuring high-order continuity constraints between the primitives. The method utilizes the Uzawa algorithm for non-linear optimization, with objective functions based on deformation energy or least squares. We show how the algorithm can be used in an interactive environment by relaxing exactness requirements while the user interactively manipulates free-form solid primitives. On current workstations, the algorithm runs in real-time for tri-quadratic volumes and close to real-time for tri-cubic volumes.

Deformations are a powerful tool for shape modeling and design. We present a new model for producing controlled spatial deformations, which we term Simple Constrained Deformations (Scodef) . The user defines a set of constraint points, giving a desired displacement and radius of influence for each. Each constraint point determines a local B-spline basis function centered at the constraint point, falling to zero for points beyond the radius. The deformed image of any point in space is a blend of these basis functions, using a projection matrix computed to satisfy the constraints. The deformation operates on the whole space regardless of the representation of the objects embedded inside the space. The constraints directly influence the final shape of the deformed objects, and this shape can be fine-tuned by adjusting the radius of influence of each constraint point. The computations required by the technique can be done very efficiently, and real-time interactive deformation editing on current workstations is possible.

Point displacement constraints constitute an attractive technique for interactive design of smooth curves, surfaces, and volumes. The user defines an arbitrary number of “control points” on the object and specifies their desired spatial location, while the system computes the object's degrees of freedom so that the constraints are satisfied. A constraint-based interface gives a feeling of direct manipulation of the object. In this article we introduce soft constraints , constraints which do not have to be met exactly. The softness of each constraint serves as a nonisotropic, local shape parameter enabling the user to explore the space of objects conforming to the constraints. Additionally, there is a global shape parameter which determines the amount of similarity of the designed object to a rest shape, or equivalently, the rigidity of the rest shape. We present an algorithm termed probabilistic point constraints (PPC) for implementing soft constraints. The PPC algorithm views constraints as stochastic measurements of the state of a static system. The softness of a constraint is derived from the covariance of the “measurement.” The resulting system of probabilistic equations is solved using the Kalman filter , a powerful estimation tool in the theory of stochastic systems. We also describe a user interface using direct-manipulation devices for specifying and visualizing covariances in 2D and 3D. The algorithm is suitable for any object represented as a parametric blend of control points, including most spline representations. The covariance of a constraint provides a continuous transition from exact interpolation to controlled approximation of the constraint. The algorithm involves only linear operations and allows real-time interactive direct manipulation of curves and surfaces on current workstations.

Abstract The same origin ray set (SORS) is a computational primitive which can be used by ray tracing, radiosity and multiple pass illumination simulation algorithms for realistic image synthesis. A SORS consists of a set of rays emanating from the same point in space. The SORS query computes the first object intersected by each ray and the intersection point. In this paper we present an efficient projection algorithm for computing a SORS query for polygonal scenes. The algorithm achieves its efficiency by separating ray‐polygon intersection detection from the computation of the intersection point between the ray and the polygon's plane. The algorithm can be integrated with all current illumination acceleration schemes. We analyse the projection algorithm and compare it to the alternative of computing the SORS query one ray at a time. The analysis' results are expressed in terms of a few intuitive parameters, measuring the success of the acceleration scheme in culling irrelevant objects and the concentration of the ray set. The projection algorithm can be up to five times more efficient, depending on these parameters and the quality of the image. The relative advantage of the projection increases with image quality.

Graphics and geometric modeling applications must provide the user with convenient methods of specifying integer and real numbers.These numbers can be coordinates of an object in some coordinate system, parameters for various geometric transformations, color and material values, and so on.A good user-interface device (UID) for number specification is an essential ingredient in any user-interface toolkit for geometric applications.Application requirements for number specification are diverse: single-and multidigit numbers, numbers in a bounded or infinite range, and both integer and real numbers.Different number bases may be desired.A desirable feature of any number specification device is the ability to define a grid and to snap the number to the nearest grid point.This is widely recognized in 2-D drawing systems and is beginning to be in 3-D systems [Bier 1990].Numbers or mouse locations are adjusted according to the current grid before being assigned to application objects.It would be intuitive to let the user control the grid parameters in the same context in which the number itself is manipulated.Current user-interface toolkits (e.g., Motif7 provide a very limited set of UIDS for number specification; usually only a text UID and a slider UID are Part of this work was done while the authors were with the Interactive Geometric Modeling group at the IBM T.

Abstract The convex differences tree (CDT) representation of a simple polygon is useful in computer graphics, computer vision, computer aided design and robotics. The root of the tree contains the convex hull of the polygon and there is a child node recursively representing every connectivity component of the set difference between the convex hull and the polygon. We give an O(n log K + K log2 n) time algorithm for constructing the CDT, where n is the number of polygon vertices and K is the number of nodes in the CDT. The algorithm is adaptive to a complexity measure defined on its output while still being worst case efficient. For simply shaped polygons, where K is a constant, the algorithm is linear. In the worst case K = O(n) and the complexity is O(n log2 n). We also give an O(n log n) algorithm which is an application of the recently introduced compact interval tree data structure.

We present a Hybrid Rendering Algorithm (HRA) for rendering parametric curves and surfaces.The algorithm uses a series of Direct Rendering Criteria (DRC) for determining whether the curve surface can be directly rendered by forward differencing with a constant step size.The DRCS test the geometric flatness of the curve/surface, its parametric uniformity, and the ability to use only integer arithmetic in the forward differencing algorithm.If any of the DRCS is not fulfilled, the curve, surface is subdivided, The location of the subdivision in parameter space is chosen to increase the chances that the new segments will satisfy the DRCS, For the integer arithmetic DRC we introduce a general method for determining an alignment of tbe forward differences.We show that for cubic [quartic) curves whose control points lie in a 128K x 128K space this alignment enables up to 2 '3(21 L) forward steps.The method is applicable to curves of any order.