arXiv subjects

Feng Ji

Publications and source records attributed to Feng Ji.

96 records · Page 6Linked to original sources

arXiv2018

Improving Multilingual Semantic Textual Similarity with Shared Sentence Encoder for Low-resource Languages

Measuring the semantic similarity between two sentences (or Semantic Textual Similarity - STS) is fundamental in many NLP applications. Despite the remarkable results in supervised settings with adequate labeling, little attention has been paid to this task in low-resource languages with insufficient labeling. Existing approaches mostly leverage machine translation techniques to translate sentences into rich-resource language. These approaches either beget language biases, or be impractical in industrial applications where spoken language scenario is more often and rigorous efficiency is required. In this work, we propose a multilingual framework to tackle the STS task in a low-resource language e.g. Spanish, Arabic , Indonesian and Thai, by utilizing the rich annotation data in a rich resource language, e.g. English. Our approach is extended from a basic monolingual STS framework to a shared multilingual encoder pretrained with translation task to incorporate rich-resource language data. By exploiting the nature of a shared multilingual encoder, one sentence can have multiple representations for different target translation language, which are used in an ensemble model to improve similarity evaluation. We demonstrate the superiority of our method over other state of the art approaches on SemEval STS task by its significant improvement on non-MT method, as well as an online industrial product where MT method fails to beat baseline while our approach still has consistently improvements.

cs.CL↗

arXiv2018

Intersections of subcomplexes in non-positively curved 2-dimensional complexes

Let $X$ be a contractible $2$-complex which is a union of two contractible subcomplexes $Y$ and $Z.$ Is the intersection $Y\cap Z$ contractible as well? In this note, we prove that the inclusion-induced map $π_{1}(Y\cap Z)\rightarrow π_{1}(Z)$ is injective if $Y$ is $π_{1}$-injective subcomplex in a locally CAT(0) 2-complex $X$. In particular, each component in the intersection of two contractible subcomplexes in a CAT(0) 2-complex is contractible.

math.GT↗

arXiv2018

Transfer Learning for Context-Aware Question Matching in Information-seeking Conversations in E-commerce

Building multi-turn information-seeking conversation systems is an important and challenging research topic. Although several advanced neural text matching models have been proposed for this task, they are generally not efficient for industrial applications. Furthermore, they rely on a large amount of labeled data, which may not be available in real-world applications. To alleviate these problems, we study transfer learning for multi-turn information seeking conversations in this paper. We first propose an efficient and effective multi-turn conversation model based on convolutional neural networks. After that, we extend our model to adapt the knowledge learned from a resource-rich domain to enhance the performance. Finally, we deployed our model in an industrial chatbot called AliMe Assist (https://consumerservice.taobao.com/online-help) and observed a significant improvement over the existing online model.

cs.CL↗

arXiv2018

Memory-augmented Dialogue Management for Task-oriented Dialogue Systems

Dialogue management (DM) decides the next action of a dialogue system according to the current dialogue state, and thus plays a central role in task-oriented dialogue systems. Since dialogue management requires to have access to not only local utterances, but also the global semantics of the entire dialogue session, modeling the long-range history information is a critical issue. To this end, we propose a novel Memory-Augmented Dialogue management model (MAD) which employs a memory controller and two additional memory structures, i.e., a slot-value memory and an external memory. The slot-value memory tracks the dialogue state by memorizing and updating the values of semantic slots (for instance, cuisine, price, and location), and the external memory augments the representation of hidden states of traditional recurrent neural networks through storing more context information. To update the dialogue state efficiently, we also propose slot-level attention on user utterances to extract specific semantic information for each slot. Experiments show that our model can obtain state-of-the-art performance and outperforms existing baselines.

cs.CL↗

arXiv2018

Partial Euler Characteristic, Normal Generations and the stable D(2) problem

We study the interplay among Wall's $D(2)$ problem, normal generation conjecture (the Wiegold Conjecture) of perfect groups and Swan's problem on partial Euler characteristic and deficiency of groups. In particular, for a 3-dimensional complex $X$ of cohomological dimension 2 with a finite fundamental group, assuming the Wiegold conjecture holds, we prove that X is homotopy equivalent to a finite 2-complex after wedging a copy of sphere $S^2$.

math.AT↗

arXiv2017

Vanishing of $L^{2}$-Betti numbers and failure of acylindrical hyperbolicity of matrix groups over rings

Let $R$ be an infinite commutative ring with identity and $n\geq 2$ be an integer. We prove that for each integer $i=0,1,\cdots ,n-2,$ the $L^{2}$-Betti number $b_{i}^{(2)}(G)=0,$ $\ $when $G=\mathrm{GL}_{n}(R)$ the general linear group, $\mathrm{SL}_{n}(R)$ the special linear group, $% E_{n}(R)$ the group generated by elementary matrices. When $R$ is an infinite principal ideal domain, similar results are obtained for $\mathrm{Sp}_{2n}(R)$ the symplectic group, $\mathrm{ESp}_{2n}(R)$ the elementary symplectic group, $\mathrm{O}(n,n)(R)$ the split orthogonal group or $\mathrm{EO}(n,n)(R)$ the elementary orthogonal group. Furthermore, we prove that $G$ is not acylindrically hyperbolic if $n\geq 4$. We also prove similar results for a class of noncommutative rings. The proofs are based on a notion of $n$-rigid rings.

math.AT↗