arXiv subjects

Benjamin Hsu

Publications and source records attributed to Benjamin Hsu.

14 recordsLinked to original sources

arXiv2024

M3T: A New Benchmark Dataset for Multi-Modal Document-Level Machine Translation

Document translation poses a challenge for Neural Machine Translation (NMT) systems. Most document-level NMT systems rely on meticulously curated sentence-level parallel data, assuming flawless extraction of text from documents along with their precise reading order. These systems also tend to disregard additional visual cues such as the document layout, deeming it irrelevant. However, real-world documents often possess intricate text layouts that defy these assumptions. Extracting information from Optical Character Recognition (OCR) or heuristic rules can result in errors, and the layout (e.g., paragraphs, headers) may convey relationships between distant sections of text. This complexity is particularly evident in widely used PDF documents, which represent information visually. This paper addresses this gap by introducing M3T, a novel benchmark dataset tailored to evaluate NMT systems on the comprehensive task of translating semi-structured documents. This dataset aims to bridge the evaluation gap in document-level NMT systems, acknowledging the challenges posed by rich text layouts in real-world applications.

cs.CL↗

arXiv2023

RAMP: Retrieval and Attribute-Marking Enhanced Prompting for Attribute-Controlled Translation

Attribute-controlled translation (ACT) is a subtask of machine translation that involves controlling stylistic or linguistic attributes (like formality and gender) of translation outputs. While ACT has garnered attention in recent years due to its usefulness in real-world applications, progress in the task is currently limited by dataset availability, since most prior approaches rely on supervised methods. To address this limitation, we propose Retrieval and Attribute-Marking enhanced Prompting (RAMP), which leverages large multilingual language models to perform ACT in few-shot and zero-shot settings. RAMP improves generation accuracy over the standard prompting approach by (1) incorporating a semantic similarity retrieval component for selecting similar in-context examples, and (2) marking in-context examples with attribute annotations. Our comprehensive experiments show that RAMP is a viable approach in both zero-shot and few-shot settings.

cs.CL↗

arXiv2023

Pseudo-Label Training and Model Inertia in Neural Machine Translation

Like many other machine learning applications, neural machine translation (NMT) benefits from over-parameterized deep neural models. However, these models have been observed to be brittle: NMT model predictions are sensitive to small input changes and can show significant variation across re-training or incremental model updates. This work studies a frequently used method in NMT, pseudo-label training (PLT), which is common to the related techniques of forward-translation (or self-training) and sequence-level knowledge distillation. While the effect of PLT on quality is well-documented, we highlight a lesser-known effect: PLT can enhance a model's stability to model updates and input perturbations, a set of properties we call model inertia. We study inertia effects under different training settings and we identify distribution simplification as a mechanism behind the observed results.

cs.CL↗

arXiv2022

MT-GenEval: A Counterfactual and Contextual Dataset for Evaluating Gender Accuracy in Machine Translation

As generic machine translation (MT) quality has improved, the need for targeted benchmarks that explore fine-grained aspects of quality has increased. In particular, gender accuracy in translation can have implications in terms of output fluency, translation accuracy, and ethics. In this paper, we introduce MT-GenEval, a benchmark for evaluating gender accuracy in translation from English into eight widely-spoken languages. MT-GenEval complements existing benchmarks by providing realistic, gender-balanced, counterfactual data in eight language pairs where the gender of individuals is unambiguous in the input segment, including multi-sentence segments requiring inter-sentential gender agreement. Our data and code is publicly available under a CC BY SA 3.0 license.

cs.CL↗

arXiv2022

Sockeye 3: Fast Neural Machine Translation with PyTorch

Sockeye 3 is the latest version of the Sockeye toolkit for Neural Machine Translation (NMT). Now based on PyTorch, Sockeye 3 provides faster model implementations and more advanced features with a further streamlined codebase. This enables broader experimentation with faster iteration, efficient training of stronger and faster models, and the flexibility to move new ideas quickly from research to production. When running comparable models, Sockeye 3 is up to 126% faster than other PyTorch implementations on GPUs and up to 292% faster on CPUs. Sockeye 3 is open source software released under the Apache 2.0 license.

cs.CL↗

arXiv2022

Contrastive Representation Learning for Cross-Document Coreference Resolution of Events and Entities

Identifying related entities and events within and across documents is fundamental to natural language understanding. We present an approach to entity and event coreference resolution utilizing contrastive representation learning. Earlier state-of-the-art methods have formulated this problem as a binary classification problem and leveraged large transformers in a cross-encoder architecture to achieve their results. For large collections of documents and corresponding set of $n$ mentions, the necessity of performing $n^{2}$ transformer computations in these earlier approaches can be computationally intensive. We show that it is possible to reduce this burden by applying contrastive learning techniques that only require $n$ transformer computations at inference time. Our method achieves state-of-the-art results on a number of key metrics on the ECB+ corpus and is competitive on others.

cs.CL↗

arXiv2022

CoCoA-MT: A Dataset and Benchmark for Contrastive Controlled MT with Application to Formality

The machine translation (MT) task is typically formulated as that of returning a single translation for an input segment. However, in many cases, multiple different translations are valid and the appropriate translation may depend on the intended target audience, characteristics of the speaker, or even the relationship between speakers. Specific problems arise when dealing with honorifics, particularly translating from English into languages with formality markers. For example, the sentence "Are you sure?" can be translated in German as "Sind Sie sich sicher?" (formal register) or "Bist du dir sicher?" (informal). Using wrong or inconsistent tone may be perceived as inappropriate or jarring for users of certain cultures and demographics. This work addresses the problem of learning to control target language attributes, in this case formality, from a small amount of labeled contrastive data. We introduce an annotated dataset (CoCoA-MT) and an associated evaluation metric for training and evaluating formality-controlled MT models for six diverse target languages. We show that we can train formality-controlled models by fine-tuning on labeled contrastive data, achieving high accuracy (82% in-domain and 73% out-of-domain) while maintaining overall quality.

cs.CL↗

arXiv2021

Hydrodynamics of quantum spin liquids

Quantum spin liquids are topological states of matter that arise in frustrated quantum magnets at low temperatures. At low energies, such states exhibit emergent gauge fields and fractionalized quasiparticles and can also possess enhanced global symmetries compared to their parent microscopic Hamiltonians. We study the consequences of this emergent gauge and symmetry structure for the hydrodynamics of quantum spin liquids. Specifically, we analyze two cases, the $U(1)$ spin liquid with a Fermi surface and the $SU(4)$-symmetric "algebraic" spin liquid. We show that the emergent degrees of freedom in the spin liquid phase lead to a variety of additional hydrodynamic modes compared to the high-temperature paramagnetic phase. We identify a hydrodynamic regime for the internal $U(1)$ gauge field common to both states, characterized by slow diffusion of the internal transverse photon.

cond-mat.str-el↗

arXiv2013

Kramers-Wannier Duality of Statistical Mechanics Applied to the Boolean Satisfiability Problem of Computer Science

We present a novel application of the Kramers-Wannier duality on one of the most important problems of computer science, the Boolean satisfiability problem (SAT). More specifically, we focus on sharp-SAT or equivalently #SAT - the problem of counting the number of solutions to a Boolean satisfaction formula. #SAT can be cast into a statistical-mechanical language, where it reduces to calculating the partition function of an Ising spin Hamiltonian with multi-spin interactions. We show that Kramers-Wannier duality can be generalized to apply to such multi-connected spin networks. We present an exact dual partner to #SAT and explicitly verify their equivalence with a few simple examples. It is shown that the NP-completeness of the original problem maps on the complexity of the dual problem of enumerating the number of non-negative solutions to a Diophantine system of equations. We discuss the implications of this duality and the prospects of similar dualities applied to computer science problems.

cond-mat.stat-mech↗

arXiv2013

Dynamical stability of the quantum Lifshitz theory in 2+1 Dimensions

The role of magnetic and electric perturbations to the quantum Lifshitz model in 2+1 dimensions are examined in this paper. The quantum Lifshitz model is an effective field theory for quantum multicritical systems, that include generalized 2D quantum dimer models in bipartite lattices and their generalizations. It describes a class of quantum phase transitions between ordered and topological phases in 2+1 dimensions. Magnetic perturbations break the dimer conservation law. Electric excitations, whose condensation lead to ordered phases, have been studied extensively both in the classical 3D model and in the quantum 2D model. However, the role of magnetic vortex excitations whose condensation drive these systems into a $\mathbb{Z}_2$ topological phase has been largely ignored. To study the interplay of both excitations, we perform a perturbative renormalization group study to one loop order and study the stability of the theory away from quantum multi criticality. This is done by generalizing the operator product expansion to anisotropic models. The relation with recent classical Monte Carlo simulations in Rokhsar-Kivelson wave functions will be discussed.

cond-mat.str-el↗

arXiv2013

The Renyi Entropy and the Multifractal Spectrum of Systems Near the Localization Transition

We show that the Rényi entropies of single particle, extended wave functions for disordered systems contain information about the multifractal spectrum. It is shown for moments of the Rényi entropy, $S_{n}$, where $|n|<1$, it is possible to extract universal information about the multifractility of such systems. This is shown through a generic calculation and then illustrated through two example models. We find good agreement between our analytic formula and numerical simulations of the two test models. Our formalism is easily extendable to generic non-interacting fermion models. It is also suggested that recent experimental advances in measuring the multifractal spectrum might allow some moments of the Rényi entropy to be measured.

cond-mat.mes-hall↗

arXiv2010

Universal Behavior of Entanglement in 2D Quantum Critical Dimer Models

We examine the scaling behavior of the entanglement entropy for the 2D quantum dimer model (QDM) at criticality and derive the universal finite sub-leading correction $γ_{QCP}$. We compute the value of $γ_{QCP}$ without approximation working directly with the wave function of a generalized 2D QDM at the Rokhsar-Kivelson QCP in the continuum limit. Using the replica approach, we construct the conformal boundary state corresponding to the cyclic identification of $n$-copies along the boundary of the observed region. We find that the universal finite term is $γ_{QCP}=\ln R-1/2$ where $R$ is the compactification radius of the bose field theory quantum Lifshitz model, the effective field theory of the 2D QDM at quantum criticality. We also demonstrated that the entanglement spectrum of the critical wave function on a large but finite region is described by the characters of the underlying conformal field theory. It is shown that this is formally related to the problems of quantum Brownian motion on $n$-dimensional lattices or equivalently a system of strings interacting with a brane containing a background electromagnetic field and can be written as an expectation value of a vertex operator.

cond-mat.stat-mech↗

arXiv2009

Quantum noise and entanglement generated by a local quantum quench

We examine the growth of entanglement under a quantum quench at point contacts of simple fractional quantum Hall fluids and its relation with the measurement of local observables. Recently Klich and Levitov proposed that the noise generated from a local quantum quench provides a measure of the entanglement entropy. Their methods were specific to non-interacting electrons and the generalization to interacting systems was left as an open question. In this work, we generalize their result to the Laughlin states. We investigate the noise generated in the current along the edge of a fractional quantum Hall state at filling factors $ν=1/m$, when a quantum point contact, initially closed, is fully opened at some initial time $t_0=0$. We find that local quenching in these systems gives time dependent correlation functions that have universal behavior on sufficiently long time and length scales. We calculate the noise and full counting statistics for $ν=1/m$ and find that in general, the entanglement entropy and noise generated are unrelated quantities. We also discuss a generalization of this problem to the critical quantum Ising spin chain.

cond-mat.mes-hall↗

arXiv2009

Universal entanglement entropy in 2D conformal quantum critical points

We study the scaling behavior of the entanglement entropy of two dimensional conformal quantum critical systems, i.e. systems with scale invariant wave functions. They include two-dimensional generalized quantum dimer models on bipartite lattices and quantum loop models, as well as the quantum Lifshitz model and related gauge theories. We show that, under quite general conditions, the entanglement entropy of a large and simply connected sub-system of an infinite system with a smooth boundary has a universal finite contribution, as well as scale-invariant terms for special geometries. The universal finite contribution to the entanglement entropy is computable in terms of the properties of the conformal structure of the wave function of these quantum critical systems. The calculation of the universal term reduces to a problem in boundary conformal field theory.

cond-mat.stat-mech↗