title stringlengths 15 188 | abstract stringlengths 400 1.8k | introduction stringlengths 9 10.5k | content stringlengths 778 41.9k | abstract_len int64 400 1.8k | intro_len int64 9 10.5k | abs_len int64 400 1.8k |
|---|---|---|---|---|---|---|
Improving Model Generalization: A Chinese Named Entity Recognition Case Study | Generalization is an important ability that helps to ensure that a machine learning model can perform well on unseen data. In this paper, we study the effect of data bias on model generalization, using Chinese Named Entity Recognition (NER) as a case study. Specifically, we analyzed five benchmarking datasets for Chine... | Named Entity Recognition (NER) is a fundamental building block for various downstream natural language processing tasks such as relation extraction Recently, by leveraging upon the pretrained language model (e.g, BERT First, we observe that in widely used Chinese NER datasets, 50% to 70% entities in test data are seen ... | In this study, we analyze five benchmarking Chinese NER datasets, including: (1) MSRA If an entity in dev/test data has been covered by the training data, we refer it as a seen entity. Otherwise, it is an unseen entity. To quantify the degree to which entities in the dev/test data have been seen in the training data, w... | 1,470 | 1,378 | 1,470 |
HOP, UNION, GENERATE: Explainable Multi-hop Reasoning without Rationale Supervision | Explainable multi-hop question answering (QA) not only predicts answers but also identifies rationales, i. e. subsets of input sentences used to derive the answers. This problem has been extensively studied under the supervised setting, where both answer and rationale annotations are given. Because rationale annotation... | Multi-hop reasoning is an important capability for any intelligent machine comprehension system. Question answering (QA) is a common application for evaluating a system's ability to reason across multiple steps Researchers have thus explored approaches that do not require rationale annotations We propose HOP, UNION, GE... | Explainable methods for multi-hop QA. Active research has been devoted to collecting human rationales for a wide range of QA tasks; a recent survey has identified 65 datasets that provide explanation annotations Other works have explored multi-hop QA with only answer supervision but not rationale supervision. As in our... | 1,094 | 1,515 | 1,094 |
MUTE: A Multimodal Dataset for Detecting Hateful Memes | The exponential surge of social media has enabled information propagation at an unprecedented rate. However, it also led to the generation of a vast amount of malign content, such as hateful memes. To eradicate the detrimental impact of this content, over the last few years hateful memes detection problem has grabbed t... | With the advent of the Internet, social media platforms (i.e., Facebook, Twitter, Instagram) significantly impact people's day-to-day life. As a result, many users communicate by posting various content in these mediums. This content includes promulgating hate speech, misinformation, aggressive and offensive views. Whi... | This work developed MUTE: a novel multimodal dataset for Bengali Hateful memes detection. The MUTE considered the memes with code-mixed and cod-switched captions. For developing the dataset, we follow the guidelines provided by For dataset construction, we have manually collected memes from various social media platfor... | 1,357 | 1,351 | 1,357 |
Large Language Models Meet Open-World Intent Discovery and Recognition: An Evaluation of ChatGPT | The tasks of out-of-domain (OOD) intent discovery and generalized intent discovery (GID) aim to extend a closed intent classifier to openworld intent sets, which is crucial to taskoriented dialogue (TOD) systems. Previous methods address them by fine-tuning discriminative models. Recently, although some studies have be... | Traditional task-oriented dialogue (TOD) systems are based on the closed-set hypothesis Previous work studied above OOD tasks by finetuning the discriminative pre-training model BERT As one of the representative LLMs, ChatGPT, developed by OpenAI, has attracted significant attention from researchers and practitioners i... | • For OOD intent discovery, ChatGPT performs far worse than the fine-tuned baselines under multi-sample or multi-category scenes, and is severely affected by the number of clusters and samples, with poor robustness. • For GID, the overall performance of ChatGPT is inferior to that of the fine-tuned baselines. The main ... | 1,109 | 1,876 | 1,109 |
Chinese NER Using Lattice LSTM | We investigate a lattice-structured LSTM model for Chinese NER, which encodes a sequence of input characters as well as all potential words that match a lexicon. Compared with character-based methods, our model explicitly leverages word and word sequence information. Compared with word-based methods, lattice LSTM does ... | As a fundamental task in information extraction, named entity recognition (NER) has received constant research attention over the recent years. The task has traditionally been solved as a sequence labeling problem, where entity boundary and category labels are jointly predicted. The current stateof-the-art for English ... | Our work is in line with existing methods using neural network for NER. How to better leverage word information for Chinese NER has received continued research attention External sources of information has been leveraged for NER. In particular, lexicon features have been widely used Lattice structured RNNs can be viewe... | 630 | 1,521 | 630 |
NLProlog: Reasoning with Weak Unification for Question Answering in Natural Language | Rule-based models are attractive for various tasks because they inherently lead to interpretable and explainable decisions and can easily incorporate prior knowledge. However, such systems are difficult to apply to problems involving natural language, due to its linguistic variability. In contrast, neural models can co... | We consider the problem of multi-hop reasoning on natural language data. For instance, consider the statements "Socrates was born in Athens" and "Athens belongs to Greece", and the question "Where was Socrates born?". There are two possible answers following from the given statements, namely "Athens" and "Greece". Whil... | Our work touches in general on weak-unification based fuzzy logic Multi-hop Reasoning for QA. One prominent approach for enabling multi-hop reasoning in neural QA models is to iteratively update a query embedding by integrating information from embeddings of context sentences, usually using an attention mechanism and s... | 1,266 | 2,160 | 1,266 |
K-best Iterative Viterbi Parsing | This paper presents an efficient and optimal parsing algorithm for probabilistic context-free grammars (PCFGs). To achieve faster parsing, our proposal employs a pruning technique to reduce unnecessary edges in the search space. The key is to repetitively conduct Viterbi inside and outside parsing, while gradually expa... | The CKY or Viterbi inside algorithm is a wellknown algorithm for PCFG parsing Despite their practical success, both pruning methods are approximate, so the solution of the parser is not always optimal, i.e., the parser does not always output the Viterbi tree. Recently, another line of work has explored A* search algo-r... | Following We assume N = {A, B, C, D}. By grouping several symbols in the same cell of the chart table, we can make a smaller table than the original one. While the original chart table in Figure . Figure symbols but also new symbols X1 and X2. The new symbols, which are made by grouping several non-terminal symbols, ar... | 685 | 1,093 | 685 |
ART: rule bAsed futuRe-inference deducTion | Deductive reasoning is a crucial cognitive ability of humanity, allowing us to derive valid conclusions from premises and observations. However, existing works mainly focus on languagebased premises and generally neglect deductive reasoning from visual observations. In this work, we introduce rule bAsed futuReinference... | Deductive reasoning is a systematic method that rigorously follows a set of explicitly given constraints (i.e., rules) to deduce valid conclusions from empirical facts through logical inferences | + Candidate Future Events: Event 1: The lady with yellow hair is putting the book on the desk into her bag. Event 2: The woman in green dress is putting down her bag. Event 3: The lady who opened the door is walking to the lamp and turning on it. stone of human psychological functioning, serving as an indispensable as... | 1,339 | 194 | 1,339 |
Evaluating and Improving the Coreference Capabilities of Machine Translation Models | Machine translation (MT) requires a wide range of linguistic capabilities, which current end-to-end models are expected to learn implicitly by observing aligned sentences in bilingual corpora. In this work, we ask: How well do MT models learn coreference resolution from implicit signal? To answer this question, we deve... | Machine translation (MT) may require coreference resolution to translate cases where the source and target language differ in their grammatical properties. For example, consider translating "The trophy didn't fit in the suitcase because it was too small" from English to French: "Le trophée ne rentrait pas dans la valis... | We start our work by extending the methodology developed in In particular, assuming a dataset of English sentences D, where each instance includes gold coreference annotation between a human entity and its pronoun (e.g., "The doctor asked the nurse to help her with the procedure."), they evaluate gender bias from Engli... | 944 | 2,238 | 944 |
Interpretable and Compositional Relation Learning by Joint Training with an Autoencoder | Embedding models for entities and relations are extremely useful for recovering missing facts in a knowledge base. Intuitively, a relation can be modeled by a matrix mapping entity vectors. However, relations reside on low dimension sub-manifolds in the parameter space of arbitrary matrices -for one reason, composition... | Broad-coverage knowledge bases (KBs) such as Freebase to predict the missing part of an incomplete triple, such as Finding Nemo, country of film, ? , by reasoning from known facts stored in the KB. As a most common approach However, modeling relations as mappings naturally requires more parameters -a general linear map... | A knowledge base (KB) is a set T of triples of the form h, r, t , where h, t ∈ E are entities and r ∈ R is a relation (e.g. The Matrix, country of film, Australia ). A relation r has its inverse r -1 ∈ R so that for every h, r, t ∈ T , we regard t, r -1 , h as also in the KB. Under this assumption and given T as traini... | 1,098 | 1,878 | 1,098 |
Finspector: A Human-Centered Visual Inspection Tool for Exploring and Comparing Biases among Foundation Models | An overview of Finspector. Users can launch Finspector in a Python notebook (e.g., Jupyter). It consists of four different sections to help users explore biases of foundation models applied to the given text: (A) users can change how (B) the distribution view of mean log probabilities are shown by selecting categories ... | Recently, pre-trained large language models (LLMs), including 'foundation models,' that are trained on large amounts of data have shown striking performances in a variety of natural language processing (NLP) tasks such as language translation, text classification, and summarization. Such models can also be fine-tuned a... | Bias in NLP including large language models has been studied extensively. Garrido-Muñoz et al. provide a survey Tenny et al. presented Language Interpretability Tool (LIT) There are several other visualization tools that help users investigate the fairness of machine learning models, primarily focusing on aspects such ... | 530 | 2,397 | 530 |
A Pre-training Strategy for Zero-Resource Response Selection in Knowledge-Grounded Conversations | Recently, many studies are emerging towards building a retrieval-based dialogue system that is able to effectively leverage background knowledge (e.g., documents) when conversing with humans. However, it is non-trivial to collect large-scale dialogues that are naturally grounded on the background documents, which hinde... | Along with the very recent prosperity of artificial intelligence empowered conversation systems in the spotlight, many studies have been focused on building human-computer dialogue systems In this paper, we consider the response selection problem in knowledge-grounded conversion and specify the background knowledge as ... | Early studies of retrieval-based dialogue focus on single-turn response selection where the input of a matching model is a message-response pair Recently, researchers pay more attention to multiturn context-response matching and usually adopt the representation-matching-aggregation paradigm to build the model. Represen... | 1,357 | 2,963 | 1,357 |
Towards Open Domain Event Trigger Identification using Adversarial Domain Adaptation | We tackle the task of building supervised event trigger identification models which can generalize better across domains. Our work leverages the adversarial domain adaptation (ADA) framework to introduce domain-invariance. ADA uses adversarial training to construct representations that are predictive for trigger identi... | Events are a key semantic phenomenon in natural language understanding. They embody a basic function of language: the ability to report happenings. Events are a basic building block for narratives across multiple domains such as news articles, stories and scientific abstracts, and are important for many downstream task... | Throughout this work, we treat the task of event trigger identification as a token-level classification task. For each token in a sequence, we predict whether it is an event trigger. To ensure that our trigger identification model can transfer across domains, we leverage the adversarial domain adaptation (ADA) framewor... | 898 | 683 | 898 |
Extracting Relational Facts by an End-to-End Neural Model with Copy Mechanism | The relational facts in sentences are often complicated. Different relational triplets may have overlaps in a sentence. We divided the sentences into three types according to triplet overlap degree, including Normal, EntityPairOverlap and SingleEn-tiyOverlap. Existing methods mainly focus on Normal class and fail to ex... | Recently, to build large structural knowledge bases (KB), great efforts have been made on extracting relational facts from natural language texts. A relational fact is often represented as a triplet which consists of two entities (an entity pair) and a semantic relation between them, such as < Chicago, country, U nited... | In this section, we introduce a differentiable neural model based on Seq2Seq learning with copy mechanism, which is able to extract multiple relational facts in an end2end fashion. Our neural model encodes a variable-length sentence into a fixed-length vector representation first and then decodes this vector into the c... | 777 | 3,709 | 777 |
Can Click Patterns across User's Query Logs Predict Answers to Definition Questions? | In this paper, we examined click patterns produced by users of Yahoo! search engine when prompting definition questions. Regularities across these click patterns are then utilized for constructing a large and heterogeneous training corpus for answer ranking. In a nutshell, answers are extracted from clicked web-snippet... | It is a well-known fact that definition queries are very popular across users of commercial search engines It is a standard practice of definition question answering (QA) systems to mine KBs (e.g., online encyclopedias and dictionaries) for reliable descriptive information on the definiendum As a means of dealing with ... | In recent years, definition QA systems have shown a trend towards the utilization of several discriminant and statistical learning techniques • centroid vector • • • Our contribution is a novel technique for obtaining heterogeneous training material for defi-nitional QA, that is to say, massive examples harvested from ... | 488 | 2,361 | 488 |
Composing Simple Image Descriptions using Web-scale N-grams | Studying natural language, and especially how people describe the world around them can help us better understand the visual world. In turn, it can also help us in the quest to generate natural language that describes this world in a human manner. We present a simple yet effective approach to automatically compose imag... | Gaining a better understanding of natural language, and especially natural language associated with images helps drive research in both computer vision and natural language processing (e.g., Our work contrasts to most previous approaches in four key aspects: first, we compose fresh sentences from scratch, instead of re... | Accommodating Uncertainty We extend candidate phrase selection in order to cope with uncertainty from the image recognition. In particular, for each object detection obj i , we include its top 3 predicted modifiers adj i1 , adj i2 , adj i3 determined by the attribute classifiers (see §2) to expand the set O 1 and O 2 a... | 803 | 1,307 | 803 |
VoiSeR: A New Benchmark for Voice-Based Search Refinement | Voice assistants, e.g., Alexa or Google Assistant, have dramatically improved in recent years. Supporting voice-based search, exploration, and refinement are fundamental tasks for voice assistants, and remain an open challenge. For example, when using voice to search an online shopping site, a user often needs to refin... | Modern voice assistants, such as Amazon Alexa or Apple Siri, make use of Natural Language Understanding (NLU) techniques to perform several tasks. Some of the most popular functions offered by these systems are based on voice-search: millions use voice assistants to access information or search for music, products or l... | In order to collect a large number of voice search refinements from multiple participants, we designed a crowdsourcing task on Amazon Mechanical Turk The design of the task was intended to make it both easy for the participants (i.e., Amazon Mechanical Turk Worker) and as realistic as possible, to provide valid linguis... | 1,582 | 2,944 | 1,582 |
On the use of Comparable Corpora to improve SMT performance | We present a simple and effective method for extracting parallel sentences from comparable corpora. We employ a statistical machine translation (SMT) system built from small amounts of parallel texts to translate the source side of the nonparallel corpus. The target side texts are used, along with other corpora, in the... | Parallel corpora have proved be an indispensable resource in Statistical Machine Translation (SMT). A parallel corpus, also called bitext, consists in bilingual texts aligned at the sentence level. They have also proved to be useful in a range of natural language processing applications like automatic lexical acquisiti... | The goal of SMT is to produce a target sentence e from a source sentence f . Among all possible target language sentences the one with the highest probability is chosen: where Pr(f |e) is the translation model and Pr(e) is the target language model (LM). This approach is usually referred to as the noisy sourcechannel a... | 613 | 2,979 | 613 |
MATINF: A Jointly Labeled Large-Scale Dataset for Classification, Question Answering and Summarization | Recently, large-scale datasets have vastly facilitated the development in nearly all domains of Natural Language Processing. However, there is currently no cross-task dataset in NLP, which hinders the development of multi-task learning. We propose MATINF, the first jointly labeled large-scale dataset for classification... | In recent years, large-scale datasets (e.g., Ima-geNet Due to the high cost of data annotation, existing NLP datasets are usually labeled for only one particular task (e.g., SQuAD In this paper, we propose Maternal and Infant Dataset (MATINF), the first large-scale dataset covering three major NLP tasks: text classific... | Topic classification is one of the most fundamental tasks in NLP. As a deeply explored task, many datasets have been used in previous research both in English (AGNews, DBPedia, Yahoo Answer However, as most of them are formal text and the target categories are general topics, even simply leveraging n-gram features coul... | 848 | 3,119 | 848 |
Discriminative Marginalized Probabilistic Neural Method for Multi-Document Summarization of Medical Literature | Although current state-of-the-art Transformerbased solutions succeeded in a wide range for single-document NLP tasks, they still struggle to address multi-input tasks such as multidocument summarization. Many solutions truncate the inputs, thus ignoring potential summary-relevant contents, which is unacceptable in the ... | The task of multi-document summarization aims to generate a compact and informative summary from a cluster of topic-related documents, which represents a very challenging natural language processing (NLP) application due to the presence of redundant and sometimes conflicting information among documents State-of-the-art... | We describe related works on multi-document summarization categorized on model architectures. Flat solutions. Flat concatenation is a simple yet powerful solution because the generation of the multi-document summary is treated as a singledocument summarization task, thus it can leverage state-of-the-art pre-trained sum... | 1,251 | 2,022 | 1,251 |
AMR Parsing as Graph Prediction with Latent Alignment | meaning representations (AMRs) are broad-coverage sentence-level semantic representations. AMRs represent sentences as rooted labeled directed acyclic graphs. AMR parsing is challenging partly due to the lack of annotated alignments between nodes in the graphs and words in the corresponding sentences. We introduce a ne... | Abstract meaning representations (AMRs) | In this work, we demonstrate that the alignments can be treated as latent variables in a joint probabilistic model and induced in such a way as to be beneficial for AMR parsing. Intuitively, in our probabilistic model, every node in a graph is assumed to be aligned to a word in a sentence: each concept is predicted bas... | 805 | 39 | 805 |
A Survey for Efficient Open Domain Question Answering | Open domain question answering (ODQA) is a longstanding task aimed at answering factual questions from a large knowledge corpus without any explicit evidence in natural language processing (NLP). Recent works have predominantly focused on improving the answering accuracy and have achieved promising progress. However, h... | Open domain question answering However, most general-purpose ODQA models are computationally intensive, slow to infer, and expensive to train. One of the reasons is the huge index/document size. For example, Towards this challenge, there are various tradeoffs in building ODQA models that meet real-world application nee... | In this section, we summarize ODQA models into three typical frameworks (see in Fig. Retriever-Reader ODQA methods generally obtain good performance. However, due to dense encoding for corpus passages and longer evidence for answer reasoning, they normally suffer from a larger index size and a slower processing speed. ... | 1,015 | 1,084 | 1,015 |
DADA: Dialect Adaptation via Dynamic Aggregation of Linguistic Rules | Existing large language models (LLMs) that mainly focus on Standard American English (SAE) often lead to significantly worse performance when being applied to other English dialects. While existing mitigations tackle discrepancies for individual target dialects, they assume access to high-accuracy dialect identificatio... | As Natural Language Processing (NLP) becomes even more impactful, the equitable distribution of its benefits becomes an increasing concern. Specifically, NLP tooling is often trained and evaluated on dominant language variants, such as Standard American English (SAE). This results in a significant decline in the perfor... | Adapter Training Feature Adapter Frozen Layer Frozen Layer Figure 2018; Existing research to mitigate this disparity has mainly focused on dialectal adaptation targeting individual dialects of interest Previous linguistic works have developed a collection of lexical and morphosyntactic features that describe the differ... | 1,046 | 516 | 1,046 |
Contrastive Learning with Keyword-based Data Augmentation for Code Search and Code Question Answering | The semantic code search is to find code snippets from the collection of candidate code snippets with respect to a user query that describes functionality. Recent work on code search proposes data augmentation of queries for contrastive learning. This data augmentation approach modifies random words in queries. When a ... | Software developers or students who major in computer science often write natural language queries to search for code snippets with desired functionality from the web search engine. The retrieved code snippets are reused or referred to improve productivity of software development. Semantic code search is a well-known c... | Some researchers proposed information retrievalbased approaches using term matching between queries and code snippets Contrastive learning encourages the distance between similar instances to be minimized and the distance between dissimilar instances to be maximized in the representation space. Recently, contrastive le... | 1,356 | 1,355 | 1,356 |
Memory-Based Dependency Parsing | This paper reports the results of experiments using memory-based learning to guide a deterministic dependency parser for unrestricted natural language text. Using data from a small treebank of Swedish, memory-based classifiers for predicting the next action of the parser are constructed. The accuracy of a classifier as... | Deterministic dependency parsing has recently been proposed as a robust and efficient method for syntactic parsing of unrestricted natural language text In this paper, we report experiments using memorybased learning The paper is structured as follows. Section 2 gives the necessary background definitions and introduces... | The linguistic tradition of dependency grammar comprises a large and fairly diverse family of theories and formalisms that share certain basic assumptions about syntactic structure, in particular the assumption that syntactic structure consists of lexical nodes linked by binary relations called dependencies (see, e.g.,... | 727 | 655 | 727 |
Language Models with Rationality | While large language models (LLMs) are proficient at question-answering (QA), it is not always clear how (or even if) an answer follows from their latent "beliefs". This lack of interpretability is a growing impediment to widespread use of LLMs. To address this, our goals are to make model beliefs and their inferential... | While large language models (LLMs) are impressive at question-answering (QA), it is not always clear how (or even if) an answer follows from their latent "beliefs" where properties of explainability, interpretability, and trust are paramount. Our goal is to help alleviate such opacity by constructing an explicit repres... | Materializing a Model's Internal Knowledge: It is now well recognized that LLMs contain extensive world knowledge with no guarantee that the generated sequence of tokens expresses the model's internal knowledge, nor entails the actual answer. Similarly, chain-ofthought (CoT) To add semantics to generations, several sys... | 1,375 | 2,256 | 1,375 |
Enhancing the generalization for Intent Classification and Out-of-Domain Detection in SLU | Intent classification is a major task in spoken language understanding (SLU). Since most models are built with pre-collected in-domain (IND) training utterances, their ability to detect unsupported out-of-domain (OOD) utterances has a critical effect in practical use. Recent works have shown that using extra data and l... | Spoken language understanding (SLU) systems play a crucial role in ubiquitous artificially intelligent voice-enabled personal assistants (PA). SLU needs to process a wide variety of user utterances and carry out user's intents, a.k.a. intent classification. Many deep neural network-based SLU models have recently been p... | In the application of intent classification, a user utterance will be either an in-domain (IND) utterance (supported by the system) or an out-of-domain (OOD) utterance (not supported by the system). The classifier is expected to correctly (1) predict the intent of supported IND utterances; and (2) detect to reject the ... | 1,095 | 1,452 | 1,095 |
AMAL: Meta Knowledge-Driven Few-Shot Adapter Learning | NLP has advanced greatly together with the proliferation of Transformer-based pre-trained language models. To adapt to a downstream task, the pre-trained language models need to be fine-tuned with a sufficient supply of annotated examples. In recent years, Adapter-based fine-tuning methods have expanded the applicabili... | Since Transformer-based In this paper, we present a cost-effective method for language model fine-tuning that is applicable, without customization, to a variety of language models and Adapter types. We focus on small to mid-sized language models such as BERT In this paper, we propose a meta-knowledgedriven few-shot ada... | Few-shot Text Classification: DS AMAL can be seen as similar with LoRA in terms of using the low-rank decomposition technique. However as a meta learning-based approach, AMAL can be applied to a broad range of language models and all existing adapter-based methods, including LoRA. We deal with the few-shot text classi... | 941 | 808 | 941 |
Neural-Symbolic Inference for Robust Autoregressive Graph Parsing via Compositional Uncertainty Quantification | Pre-trained seq2seq models excel at graph semantic parsing with rich annotated data, but generalize worse to out-of-distribution (OOD) and long-tail examples. In comparison, symbolic parsers under-perform on populationlevel metrics, but exhibit unique strength in OOD and tail generalization. In this work, we study comp... | A structured account of compositional meaning has become a longstanding goal for Natural Language Processing. To this end, a number of efforts have focused on encoding semantic relationships and attributes into graph-based meaning representations (MRs, see Appendix A for details). In particular, graph semantic parsing ... | In this work, we take the representations from English Resource Grammar (ERG; ERG can be presented into different types of annotation formalism In this section, we review the state-of-the-art symbolic and neural parsers utilized in our work, i.e., the ACE parser We hypothesize that when the neural seq2seq model is unce... | 1,299 | 4,583 | 1,299 |
CREST: A Joint Framework for Rationalization and Counterfactual Text Generation | Selective rationales and counterfactual examples have emerged as two effective, complementary classes of interpretability methods for analyzing and training NLP models. However, prior work has not explored how these methods can be integrated to combine their complementary advantages. We overcome this limitation by intr... | As NLP models have become larger and less transparent, there has been a growing interest in developing methods for finer-grained interpretation and control of their predictions. One class of methods leverages selective rationalization as improved robustness to input perturbations This paper is motivated by the observat... | The traditional framework of rationalization involves training two components cooperatively: the generator-which consists of an encoder and an explainer-and the predictor. The generator encodes the input and produces a "rationale" (e.g., word highlights), while the predictor classifies the text given only the rationale... | 1,301 | 1,578 | 1,301 |
Improving Speech Translation by Understanding and Learning from the Auxiliary Text Translation Task | Pretraining and multitask learning are widely used to improve the speech to text translation performance. In this study, we are interested in training a speech to text translation model along with an auxiliary text to text translation task. We conduct a detailed analysis to understand the impact of the auxiliary task o... | End-to-end methods have achieved significant progress in speech to text translation (ST) and even surpassed the traditional pipeline-based methods in some applications In this study, we focus on training the ST model along with an auxiliary text to text machine translation (MT) task. We are interested in the task inter... | The proposed ST system is co-trained with the MT task as depicted in Figure The model has two encoders, a text encoder and a speech encoder, to take text and speech input respectively. The decoder is shared between the two tasks. To encourage knowledge sharing between the two tasks, the top encoder layers are also shar... | 1,627 | 2,886 | 1,627 |
Learning the Visualness of Text Using Large Vision-Language Models | Visual text evokes an image in a person's mind, while non-visual text fails to do so. A method to automatically detect visualness in text will enable text-to-image retrieval and generation models to augment text with relevant images. This is particularly challenging with long-form text as text-to-image generation and r... | People typically communicate knowledge and information textually, but most prefer to consume visually rich content. Text-to-image generation/retrieval models could augment text with appropriate images, aiding the creation of appealing and easy-to-understand documents. Models like DALL-E Prior approaches for quantifying... | Fine-tuning vision-language models for downstream tasks: Large vision-language models like CLIP Our proposed fine-tuning approach follows multistage training of a large vision-language model CLIP The formulation of the training objective (discussed later) requires positive examples comprising vi- Example text from TIME... | 1,366 | 1,885 | 1,366 |
The lack of theory is painful: Modeling Harshness in Peer Review Comments | The peer-review system has primarily remained the central process of all science communications. However, research has shown that the process manifests a power-imbalance scenario where the reviewer enjoys a position where their comments can be overly critical and wilfully obtuse without being held accountable. This bri... | The peer-review system has largely remained the central and universal quality control system in all scientific fields. "The very act of evaluating another's work is a thinly disguised instructional relationship of authority; an inherently unequal interaction because the power to criticise is non-reciprocal and lies exc... | There is a growing body of literature on Natural Language Processing (NLP) for peer reviews and scientific literature in general. For example, datasets like PeerRead We define review harshness as a metric encompassing two orthogonal dimensions. The first dimension concerns the evaluative focus of the comment, and the s... | 1,054 | 2,950 | 1,054 |
Exact Hard Monotonic Attention for Character-Level Transduction | Many common character-level, string-tostring transduction tasks, e.g. graphemeto-phoneme conversion and morphological inflection, consist almost exclusively of monotonic transduction. Neural sequence-tosequence models with soft attention, which are non-monotonic, often outperform popular monotonic models. In this work,... | Many tasks in natural language can be treated as character-level, string-to-string transduction. The current dominant method is the neural sequenceto-sequence model with soft attention The standard versions of both soft and hard attention are non-monotonic. However, if we look at the data in grapheme-to-phoneme convers... | We assume the source string x ∈ Σ * x and the target string y ∈ Σ * y have finite vocabularies Σ x = {x 1 , . . . , x |Σx| } and Σ y = {y 1 , . . . , y |Σy| }, respectively. In tasks where the tag is provided, i.e., labeled transduction Hard attention was first introduced to the literature by where we show how one can ... | 876 | 591 | 876 |
How to Enhance Causal Discrimination of Utterances: A Case on Affective Reasoning | Our investigation into the Affective Reasoning in Conversation (ARC) task highlights the challenge of causal discrimination. Almost all existing models, including large language models (LLMs), excel at capturing semantic correlations within utterance embeddings but fall short in determining the specific causal relation... | Nowadays, numerous conversation recognition tasks (such as Emotion Recognition in Conversation (ERC) task However, when it comes to the relationship between two utterances, denoted as A and B, wherein their embeddings can be fitted, various possible relationships exist: A acts as the cause of B (A → B), A acts as the o... | For notational consistency, we use the following terminology. The target utterance U t is the t th utterances of a conversation D = (U 1 , U 2 , U 3 , . . . , U N ) where N is the maximum number of utterances in this conversation and 0 < t ⩽ N . The emotion label Emo t denotes the emotion type of U t . The emotion-caus... | 1,033 | 2,641 | 1,033 |
Character-Level Translation with Self-attention | We explore the suitability of self-attention models for character-level neural machine translation. We test the standard transformer model, as well as a novel variant in which the encoder block combines information from nearby characters using convolutions. We perform extensive experiments on WMT and UN datasets, testi... | Most existing Neural Machine Translation (NMT) models operate on the word or subword-level, which tends to make these models memory inefficient because of large vocabulary sizes. Characterlevel models In this work, we perform an in-depth investigation of the suitability of self-attention models for character-level tran... | Fully character-level translation was first tackled in Multilingual training of character-level models is possible not only for languages that have almost identical character vocabularies, such as French and Spanish, but even for distant languages that can be mapped to a common character-level vocabulary, for example, ... | 618 | 1,123 | 618 |
Which Melbourne? Augmenting Geocoding with Maps | The purpose of text geolocation is to associate geographic information contained in a document with a set (or sets) of coordinates, either implicitly by using linguistic features and/or explicitly by using geographic metadata combined with heuristics. We introduce a geocoder (location mention disambiguator) that achiev... | Geocoding | Depending on the task objective, geocoding methodologies can be divided into two distinct categories: (1) document geocoding, which aims at locating a piece of text as a whole, for example geolocating Twitter users Computational methods in geocoding broadly divide into rule-based, statistical and machine learning-based... | 1,004 | 9 | 1,004 |
Every word counts: A multilingual analysis of individual human alignment with model attention | Human fixation patterns have been shown to correlate strongly with Transformer-based attention. Those correlation analyses are usually carried out without taking into account individual differences between participants and are mostly done on monolingual datasets making it difficult to generalise findings. In this paper... | Recent research has shown that relative importance metrics in neural language models correlate strongly with human attention, i.e., fixation durations extracted from eye-tracking recordings during reading In this short paper, we approach this by analysing (i) differences in correlation between machine attention and hum... | We quantify the individual differences in human alignment with Transformer-based attention in a correlation study where we compare relative fixation duration from native speakers of 13 different languages on the MECO corpus The results show that (i) the correlation varies greatly across languages, (ii) L1 reading data ... | 754 | 1,312 | 754 |
PEARL: Prompting Large Language Models to Plan and Execute Actions Over Long Documents | Strategies such as chain-of-thought prompting improve the performance of large language models (LLMs) on complex reasoning tasks by decomposing input examples into intermediate steps. However, it remains unclear how to apply such methods to reason over long input documents, in which both the decomposition and the outpu... | Performing complex reasoning over long input documents often requires forming high-level abstractions of the text (e.g., plots and themes in a narrative) and then conducting a variety of inferences on top of those abstractions Mine helpful actions from training set questions DEFINE(X), COMPARE(X,Y), FIND_EMOTION(X),... | Execute the plan step-by-step Plan Execution open_conv = "In the initial conversation, Phil Conover is excited about his upcoming mission to be the first man to see the other side of the moon ...." Question: What part of the final scene best connects to the story's opening conversation? 1.open_conv = FIND_ELEMENT(CTX,"... | 1,305 | 320 | 1,305 |
Unsupervised Joint Training of Bilingual Word Embeddings | State-of-the-art methods for unsupervised bilingual word embeddings (BWE) train a mapping function that maps pre-trained monolingual word embeddings into a bilingual space. Despite its remarkable results, unsupervised mapping is also well-known to be limited by the dissimilarity between the original word embedding spac... | Bilingual word embeddings (BWE) represent the vocabulary of two languages in one common continuous vector space. They are known to be useful in a wide range of cross-lingual NLP tasks. The most prevalent methods for training BWE are so-called mapping methods In spite of their success, unsupervised mapping methods are i... | On the strong assumption that existing algorithms for joint training of BWE are robust enough even with very noisy parallel training data, we formulate the following research question: Do synthetic sentence pairs supply useful bilingual contextual information for learning better BWE? Previous work on joint training of ... | 922 | 969 | 922 |
CONQRR: Conversational Query Rewriting for Retrieval with Reinforcement Learning | Compared to standard retrieval tasks, passage retrieval for conversational question answering (CQA) poses new challenges in understanding the current user question, as each question needs to be interpreted within the dialogue context. Moreover, it can be expensive to retrain well-established retrievers such as search e... | Passage retrieval in an open-domain conversational question answering (CQA) system Therefore, in this paper, we focus on query rewriting for the task of conversational passage retrieval in a CQA dialogue with any off-the-shelf retrieval system that can only be used as a black box. Specifically, we seek to build a QR mo... | Most existing CQA datasets In contrast, QReCC before generating an answer to the question. A few recent works Most existing conversational retrieval models require fine-tuning a retriever of a specific type (Table Query Rewriting (QR) In order to directly use an off-the-shelf retriever as we aim to do, conversational Q... | 1,025 | 1,973 | 1,025 |
The (Non-)Utility of Structural Features in BiLSTM-based Dependency Parsers | Classical non-neural dependency parsers put considerable effort on the design of feature functions. Especially, they benefit from information coming from structural features, such as features drawn from neighboring tokens in the dependency tree. In contrast, their BiLSTM-based successors achieve state-ofthe-art perform... | When designing a conventional non-neural parser substantial effort is required to design a powerful feature extraction function. Such a function Recently, Since the introduction of the K&G architecture BiLSTM-based parsers have become standard in the field. Inspired by recent work | Our graph-and transition-based parsers are based on the K&G architecture (see Figure In both transition-and graph-based architectures input tokens are represented in the same way (see level The embeddings are initialized randomly at training time and trained together with the model. The representations x i encode words... | 963 | 281 | 963 |
Accelerating Sparse Matrix Operations in Neural Networks on Graphics Processing Units | Graphics Processing Units (GPUs) are commonly used to train and evaluate neural networks efficiently. While previous work in deep learning has focused on accelerating operations on dense matrices/tensors on GPUs, efforts have concentrated on operations involving sparse data structures. Operations using sparse structure... | The speedups introduced by parallel architectures inspired the development of accelerators tailored towards specialized functions. Graphics Processing Units (GPUs) are now a standard platform for deep learning. GPUs provide faster model training and inference times compared to serial processors, because they can parall... | GPUs are widely used to accelerate a variety of non-neural tasks such as search CPUs call special functions, also called kernels, to execute a set of instructions in parallel using multiple threads on the GPU. Kernels can be configured to create and execute an arbitrary number of threads. The threads in a kernel are gr... | 923 | 1,564 | 923 |
Enhancing Air Quality Prediction with Social Media and Natural Language Processing | Accompanied by modern industrial developments, air pollution has already become a major concern for human health. Hence, air quality measures, such as the concentration of PM 2.5 , have attracted increasing attention. Even some studies apply historical measurements into air quality forecast, the changes of air quality ... | In recent centuries, industrialization has considerably changed human society by providing a stimulus to economic growth and improved life quality. However, the advancement is accompanied by the increase in air pollutant emissions and risks to public health. As a consequence, predicting real-time air quality informatio... | Following the previous studies In most of the cities, the majority of tweets should be irrelevant to air quality because users are less likely to discuss air quality situations unless there is a dramatic change. Hence, we need to filter tweets before using them for air quality prediction. Following the previous work (S... | 1,050 | 1,857 | 1,050 |
AUTONLU: An On-demand Cloud-based Natural Language Understanding System for Enterprises | With the renaissance of deep learning, neural networks have achieved promising results on many natural language understanding (NLU) tasks. Even though the source codes of many neural network models are publicly available, there is still a large gap from open-sourced models to solving real-world problems in enterprises.... | In recent years, many deep learning methods have achieved impressive results on a wide range of tasks, ranging from question answering • Ease of use. AUTONLU aims to help users with limited technical knowledge to train and test models on their datasets. We provide GUI modules to accommodate the most common use-cases, f... | Closely related branches of work to ours are toolkits and frameworks designed to provide a suite of state-of-the-art NLP models to users In 2018, Google introduced AutoML Natural Language 1 , a platform that enables users to build and deploy machine learning models for various NLP tasks. Our system is different from Au... | 1,047 | 1,416 | 1,047 |
Inference Rules and their Application to Recognizing Textual Entailment | In this paper, we explore ways of improving an inference rule collection and its application to the task of recognizing textual entailment. For this purpose, we start with an automatically acquired collection and we propose methods to refine it and obtain more rules using a hand-crafted lexical resource. Following this... | Textual inference plays an important role in many natural language processing (NLP) tasks. In recent years, the recognizing textual entailment (RTE) Studies such as A typical example is the following RTE pair in which accelerate to in H is used as an alternative formulation for reach speed of in T. T: The high-speed tr... | A number of automatically acquired inference rule/paraphrase collections are available, such as The DIRT algorithm has been introduced by An inference rule in DIRT is a pair of binary relations pattern 1 (X, Y ), pattern 2 (X, Y ) which stand in an inference relation. pattern 1 and pattern 2 are chains in dependency tr... | 625 | 2,386 | 625 |
Type Level Clustering Evaluation: New Measures and a POS Induction Case Study | Clustering is a central technique in NLP. Consequently, clustering evaluation is of great importance. Many clustering algorithms are evaluated by their success in tagging corpus tokens. In this paper we discuss type level evaluation, which reflects class membership only and is independent of the token statistics of a p... | Clustering is a central machine learning technique. In NLP, clustering has been used for virtually every semi-and unsupervised task, including POS tagging In this paper we discuss type level evaluation, which evaluates the set membership structure created by the clustering, independently of the token statistics of the ... | This section motivates why both type and token level external evaluations should be done, even in the monosemous case. Clustering algorithms compute a set of induced clusters (a clustering). Some algorithms directly compute a clustering, while some others produce a tagging of corpus tokens from which a clustering can b... | 490 | 1,601 | 490 |
Integrating Language Models into Direct Speech Translation: An Inference-Time Solution to Control Gender Inflection | When translating words referring to the speaker, speech translation (ST) systems should not resort to default masculine generics nor rely on potentially misleading vocal traits. Rather, they should assign gender according to the speakers' preference. The existing solutions to do so, though effective, are hardly feasibl... | The problem of gender bias in automatic translation particularly emerges when translating from genderless or notional gender languages (e.g., English) -which feature limited gender-specific markinginto grammatical gender languages (e.g., Spanish) -which exhibit a rich lexical and morpho-syntactic system of gender So fa... | The autoregressive decoder of an encoder-decoder architecture is trained to predict the next target token given the previous ones and the encoder output. Thereby, it implicitly learns to model the target language from the training data, thus developing an internal language model (ILM) The integration of end-to-end mode... | 1,021 | 1,254 | 1,021 |
Beyond Accuracy: Behavioral Testing of NLP Models with CheckList | Although measuring held-out accuracy has been the primary approach to evaluate generalization, it often overestimates the performance of NLP models, while alternative approaches for evaluating models either focus on individual tasks or on specific behaviors. Inspired by principles of behavioral testing in software engi... | One of the primary goals of training NLP models is generalization. Since testing "in the wild" is expensive and does not allow for fast iterations, the standard paradigm for evaluation is using trainvalidation-test splits to estimate the accuracy of the model, including the use of leader boards to track progress on a t... | Conceptually, users "CheckList" a model by filling out cells in a matrix (Figure While testing individual components is a common practice in software engineering, modern NLP models are rarely built one component at a time. Instead, CheckList encourages users to consider how different natural language capabilities are m... | 1,044 | 677 | 1,044 |
FAD-X: Fusing Adapters for Cross-lingual Transfer to Low-Resource Languages | Adapter-based tuning, by adding light-weight adapters to multilingual pretrained language models (mPLMs), selectively updates language-specific parameters to adapt to a new language, instead of finetuning all shared weights. This paper explores an effective way to leverage a public pool of pretrained language adapters,... | While fine-tuning the multilingual pretrained language models (mPLMs), such as mBERT To overcome this challenge, MAD-X language-and task-specific parameters, which can also be released as pretrained adapters. However, we argue that a significant resource imbalance yet remains, especially for LRLs. To illustrate, Figure... | We first briefly review MAD-X To overcome the lack of resources for LA/TA observed for LRLs, we propose FAD-X. Our key idea is fusing task adapters trained with pretrained adapters in other languages. More formally, given a pool of n pretrained adapters, L = {LA l 1 , • • • , LA ln }, our goal is fusing T A l i trained... | 928 | 1,453 | 928 |
Learning to Discover, Ground and Use Words with Segmental Neural Language Models | We propose a segmental neural language model that combines the generalization power of neural networks with the ability to discover word-like units that are latent in unsegmented character sequences. In contrast to previous segmentation models that treat word segmentation as an isolated task, our model unifies word dis... | How infants discover words that make up their first language is a long-standing question in developmental psychology In this paper, we introduce a single model that discovers words, learns how they fit together (not just locally, but across a complete sentence), and grounds them in learned representations of naturalist... | We now describe the segmental neural language model (SNLM). Refer to Figure The SNLM defines the distribution over x as the marginal distribution over all segmentations that give rise to x, i.e., p(x) = s:π(s)=x p(s). (1) To define the probability of p(s), we use the chain rule, rewriting this in terms of a product of ... | 787 | 3,910 | 787 |
A Joint Neural Model for Information Extraction with Global Features | Most existing joint neural models for Information Extraction (IE) use local task-specific classifiers to predict labels for individual instances (e.g., trigger, relation) regardless of their interactions. For example, a VICTIM of a DIE event is likely to be a VICTIM of an AT-TACK event in the same sentence. In order to... | Information Extraction (IE) aims to extract structured information from unstructured texts. It is a complex task comprised of a wide range of subtasks, such as named, nominal, and pronominal mention extraction, entity linking, entity coreference resolution, relation extraction, event extraction, and event coreference r... | Example: Prime Minister Abdullah Gul resigned earlier Tuesday to make way for Erdogan, who won a parliamentary seat in by-elections Sunday. person To address this issue, we propose a joint neu- ral framework, ONEIE, to perform end-to-end IE with global constraints. As Figure To the best of our knowledge, ONEIE is the f... | 632 | 391 | 632 |
BiSET: Bi-directional Selective Encoding with Template for Abstractive Summarization | The success of neural summarization models stems from the meticulous encodings of source articles. To overcome the impediments of limited and sometimes noisy training data, one promising direction is to make better use of the available training data by applying filters during summarization. In this paper, we propose a ... | Abstractive summarization aims to shorten a source article or paragraph by rewriting while preserving the main idea. Due to the difficulties in rewriting long documents, a large body of research on this topic has focused on paragraph-level article summarization. Among them, sequence-tosequence models have become the ma... | Our framework includes three key modules: Retrieve, Fast Rerank, and BiSET. For each source article, Retrieve aims to return a few candidate templates from the training corpus. Then, the Fast Rerank module quickly identifies a best template from the candidates. Finally, BiSET mutually selects important information from... | 762 | 1,071 | 762 |
Language Models as Inductive Reasoners | Inductive reasoning is a core component of human intelligence. In the past research of inductive reasoning within computer science, formal language is used as representations of knowledge (facts and rules, more specifically). However, formal language can cause systematic problems for inductive reasoning such as disabil... | Inductive reasoning is to reach to a hypothesis (usually a rule that explains an aspect of the law of nature) based on pieces of evidence (usually observed facts of the world), where the observations can not provide conclusive support to the hypothesis that the hypothesis supports more than mere reformulation of the co... | The Venus flytrap is a carnivorous plant native to subtropical wetlands on the East Coast of the United States in North Carolina and South Carolina. It catches its prey-chiefly insects and arachnids-with a trapping structure formed by the terminal portion of each of the plant's leaves, which is triggered by tiny hairs ... | 1,490 | 867 | 1,490 |
Life is a Circus and We are the Clowns: Automatically Finding Analogies between Situations and Processes | Analogy-making gives rise to reasoning, abstraction, flexible categorization and counterfactual inference -abilities lacking in even the best AI systems today. Much research has suggested that analogies are key to non-brittle systems that can adapt to new domains. Despite their importance, analogies received little att... | The ability to find parallels across diverse domains and transfer ideas across them is one of the pinnacles of human cognition. The analogous reasoning process allows us to abstract information, form flexible concepts and solve problems based on our previous experience Surprisingly, despite analogy's important role in ... | Our framework is based on Gentner's structure mapping theory (SMT) Intuitively, we want similarity to be high if the two sets share many distinct relations. For example, {provide, destroy}, should be more similar to {supply, ruin} than to {destroy, ruin} as the last set does not include any relation similar to provide.... | 916 | 1,838 | 916 |
Interventional Training for Out-Of-Distribution Natural Language Understanding | Out-of-distribution (OOD) settings are used to measure a model's performance when the distribution of the test data is different from that of the training data. NLU models are known to suffer in OOD settings (Utama et al., 2020b). We study this issue from the perspective of causality, which sees confounding bias as the... | From the era of word embeddings Recently, causal inference has been adopted in NLP to identify robust correlations by analyzing reliable causal effects between variables (Zhang A common solution of deconfounding is intervention However, the confounder C is not always observed. Furthermore, confounders can be multifacto... | Causal Intervention is the core idea of this paper. We formulate NLU tasks with a causal graph Naïve model training, i.e., empirical risk minimization (ERM) where the bias is introduced via P (C|X). For example, consider the NLI task. Let X be a pair of two sentences (premise and hypothesis) and Y the entailment label.... | 973 | 1,811 | 973 |
Food Knowledge Representation Learning with Adversarial Substitution | Knowledge graph embedding (KGE) has been well-studied in general domains, but has not been examined for food computing. To fill this gap, we perform knowledge representation learning over a food knowledge graph (KG). We employ a pre-trained language model to encode entities and relations, thus emphasizing contextual in... | Structured knowledge furnishes an in-depth understanding of the world. Knowledge graph embedding (KGE) maps entities and relations into vectors while retaining their semantics domain-specific KGs As for encoding models in KGE, most deep learning-based methods like convolutional neural networks (CNN) Large-scale food da... | The models that encode the interactions of entities and relations in knowledge graphs can be categorized into: linear/bilinear models, factorization models, and neural networks. Among the neural networks-based models, Convolutional Neural Networks (CNNs) are utilized for learning deep expressive features Previous work ... | 1,360 | 3,350 | 1,360 |
Redistributing Low-Frequency Words: Making the Most of Monolingual Data in Non-Autoregressive Translation | Knowledge distillation (KD) is the preliminary step for training non-autoregressive translation (NAT) models, which eases the training of NAT models at the cost of losing important information for translating low-frequency words. In this work, we provide an appealing alternative for NAT -monolingual KD, which trains NA... | Non-autoregressive translation (NAT, Although the standard KD on original bilingual data eases the training of NAT models, distillation may lose some important information in the raw training data, leading to more errors on predicting low-frequency words Specifically, we leverage the monolingual data to perform KD (mon... | Non-Autoregressive Translation Recent years have seen a surge of interest in NAT Standard Knowledge Distillation Knowledge distillation is the preliminary step for training NAT models by reducing the modes in the original bilingual data, which makes NAT easily acquire more deterministic knowledge and achieve significan... | 1,430 | 2,160 | 1,430 |
A Walk-based Model on Entity Graphs for Relation Extraction | We present a novel graph-based neural network model for relation extraction. Our model treats multiple pairs in a sentence simultaneously and considers interactions among them. All the entities in a sentence are placed as nodes in a fully-connected graph structure. The edges are represented with position-aware contexts... | Relation extraction (RE) is a task of identifying typed relations between known entity mentions in a sentence. Most existing RE models treat each relation in a sentence individually Multiple relations in a sentence between entity mentions can be represented as a graph. Neural graph-based models have shown significant i... | The goal of the RE task is given a sentence, entity mentions and their semantic types, to extract and classify all related entity pairs (target pairs) in the sentence. The proposed model consists of five stacked layers: embedding layer, BLSTM Layer, edge representation layer, walk aggregation layer and finally a classi... | 732 | 1,161 | 732 |
FETA: A Benchmark for Few-Sample Task Transfer in Open-Domain Dialogue | Task transfer, transferring knowledge contained in related tasks, holds the promise of reducing the quantity of labeled data required to finetune language models. Dialogue understanding encompasses many diverse tasks, yet task transfer has not been thoroughly studied in conversational AI. This work explores conversatio... | Improving sample efficiency through transfer learning has been a long-standing challenge in the machine learning and natural language processing communities Two essential transfer learning settings, namely domain adaptation and task transfer, have been studied on language tasks Prior studies have focused on cross-datas... | Transfer Learning in NLP Prior works on transfer learning in NLP have studied a wide variety of topics, including domain adaptation More recently, DialoGLUE In this section, we briefly define intra-dataset task transfer, the problem setting of FETA. Then, we introduce FETA, our benchmark for few-sample task transfer in... | 1,262 | 2,072 | 1,262 |
A Multitask Learning Approach for Diacritic Restoration | In many languages like Arabic, diacritics are used to specify pronunciations as well as meanings. Such diacritics are often omitted in written text, increasing the number of possible pronunciations and meanings for a word. This results in a more ambiguous text making computational processing on such text more difficult... | In contrast to English, some vowels in languages such as Arabic and Hebrew are not part of the alphabet and diacritics are used for vowel specification. Diacritic restoration (or diacritization) is the process of restoring these missing diacritics for every character in the written texts. It can specify pronunciation a... | We formulate the problem of (full) diacritic restoration (DIAC) as follows: given a sequence of characters, we identify the diacritic corresponding to each character in that sequence from the following set of diacritics {a, u, i, o, K, F, N, ∼, ∼a, ∼u, ∼i, ∼F, ∼K, and ∼N}. We additionally consider three auxiliary tasks... | 1,232 | 1,967 | 1,232 |
Story Centaur: Large Language Model Few Shot Learning as a Creative Writing Tool | Few shot learning with large language models has the potential to give individuals without formal machine learning training the access to a wide range of text to text models. We consider how this applies to creative writers and present STORY CENTAUR, a user interface for prototyping few shot models and a set of recombi... | One of the most promising possibilities for large language models (LLMs) is few-shot learning We present STORY CENTAUR, a Human-Computer Interface that closes the gap between non-technical users and the power and possibilities of few shot learning, with the intended audience of writers of creative text. It is our inten... | The observation that simple "fill-in-the-blank" neural network models trained on large quantities of text can be used for problems beyond their primary learning objective dates back to word2vec Representation learning techniques made steady advances, expanding to sentence level contextually sensitive word embedding wit... | 716 | 677 | 716 |
Towards Fine-grained Text Sentiment Transfer | In this paper, we focus on the task of finegrained text sentiment transfer (FGST). This task aims to revise an input sequence to satisfy a given sentiment intensity, while preserving the original semantic content. Different from conventional sentiment transfer task that only reverses the sentiment polarity (positive/ne... | Text sentiment transfer aims to rephrase the input to satisfy a given sentiment label (value) while preserving its original semantic content. It facilitates various NLP applications, such as automatically converting the attitude of review and fighting against offensive language in social media (dos Figure positive and ... | Given an input sequence x and a target sentiment intensity value v y , the FTST task aims to generate a sequence y which not only expresses the target sentiment intensity v y , but also preserve the original semantic content of the input x. Without loss of generality, we limit the sentiment intensity value v y ranging ... | 466 | 2,563 | 466 |
Word Error Rate Estimation for Speech Recognition: e-WER | Measuring the performance of automatic speech recognition (ASR) systems requires manually transcribed data in order to compute the word error rate (WER), which is often time-consuming and expensive. In this paper, we propose a novel approach to estimate WER, or e-WER, which does not require a gold-standard transcriptio... | Automatic Speech Recognition (ASR) has made rapid progress in recent years, primarily due to advances in deep learning and powerful computing platforms. As a result, the quality of ASR has improved dramatically, leading to various applications, such as speech-to-speech translation, personal assistants, and broadcast me... | Estimating the probability of error of each word in a recognised word sequence has been successfully used to detect insertions, substitutions, and interword deletions In our framework, we use two speech recognition systems; a word-based LVCSR system and a grapheme-sequence based system. Following Our model is required ... | 808 | 2,335 | 808 |
Inferring semantic roles using sub-categorization frames and maximum entropy model | In this paper, we propose an approach for inferring semantic role using subcategorization frames and maximum entropy model. Our approach aims to use the sub-categorization information of the verb to label the mandatory arguments of the verb in various possible ways. The ambiguity between the assignment of roles to mand... | Semantic role labelling is the task of assigning appropriate semantic roles to the arguments of a verb. The semantic role information is important for various applications in NLP such as Machine Translation, Question Answering, Informa-tion Extraction etc. In general, semantic role information is useful for sentence un... | The following are the features used to train the maximum entropy classifier for both the argument identification and argument classification. We used only simple features for these experiments, we are planning to use richer features in the near future. 1. Verb/Predicate. 6. The path of the constituent to the verb phras... | 918 | 2,618 | 918 |
TBL-Improved Non-Deterministic Segmentation and POS Tagging for a Chinese Parser | Although a lot of progress has been made recently in word segmentation and POS tagging for Chinese, the output of current state-of-the-art systems is too inaccurate to allow for syntactic analysis based on it. We present an experiment in improving the output of an off-the-shelf module that performs segmentation and tag... | Word segmentation and tagging are the necessary initial steps for almost any language processing system, and Chinese parsers are no exception. However, automatic Chinese word segmentation and tagging has been recognized as a very difficult task Second, in addition to the two problems described above, segmentation and t... | The improved tokenizer-tagger we developed is part of a larger system, namely a deep Chinese grammar (3) The output of the Chinese LFG consists of a Constituent Structure (c-structure) and a Functional Structure (f-structure) for each sentence. While c-structure represents phrasal structure and linear word order, f-str... | 1,042 | 2,284 | 1,042 |
Knowing the No-match: Entity Alignment with Dangling Cases | This paper studies a new problem setting of entity alignment for knowledge graphs (KGs). Since KGs possess different sets of entities, there could be entities that cannot find alignment across them, leading to the problem of dangling entities. As the first attempt to this problem, we construct a new dataset and design ... | Knowledge graphs (KGs) have evolved to be the building blocks of many intelligent systems Nonetheless, to practically support the alignment of KGs as a real-world task, existing studies suffer one common problem of identifying entities without alignment across KGs (called dangling entities). Specifically, current metho... | We hereby describe the problem setting of our task and introduce the new dataset. A KG is a set of relational triples T ⊆ E × R × E, where E and R denote vocabularies of entities and relations, respectively. Without loss of generality, we consider entity alignment between two KGs, i.e., a source KG Given a small set of... | 1,161 | 2,330 | 1,161 |
Improving Self-training for Cross-lingual Named Entity Recognition with Contrastive and Prototype Learning | In cross-lingual named entity recognition (NER), self-training is commonly used to bridge the linguistic gap by training on pseudolabeled target-language data. However, due to sub-optimal performance on target languages, the pseudo labels are often noisy and limit the overall performance. In this work, we aim to improv... | Cross-lingual named entity recognition (NER) To optimize self-training for cross-lingual NER, several methods have been proposed to improve the quality of pseudo labels. One line of work focuses on selecting curated pseudo-labeled data for selftraining via reinforcement learning In this work, we take a different approa... | Cross-lingual named entity recognition aims to train a NER model with labeled data in a source language, and evaluate it on test data in target languages. Following previous works Following Typically, self-training (or teacher-student learning) for cross-lingual NER first trains a teacher model M(θ t ) on the available... | 1,199 | 2,092 | 1,199 |
QUEST: A Retrieval Dataset of Entity-Seeking Queries with Implicit Set Operations | Formulating selective information needs results in queries that implicitly specify set operations, such as intersection, union, and difference. For instance, one might search for "shorebirds that are not sandpipers" or "science-fiction films shot in England". To study the ability of retrieval systems to meet such infor... | People often express their information needs with multiple preferences or constraints. Queries corresponding to such needs typically implicitly express set operations such as intersection, difference, and union. For example, a movie-goer might be looking for a science-fiction film from the 90s which does not feature al... | Previous work in question answering and information retrieval has focused on QA over knowledge bases as well as open-domain QA and retrieval over a set of entities or documents. We highlight how these relate to our work below. Knowledge Base QA Several datasets have been proposed for question answering over knowledge b... | 1,280 | 3,970 | 1,280 |
SUMMEDITS: Measuring LLM Ability at Factual Reasoning Through The Lens of Summarization | With the recent appearance of LLMs in practical settings, having methods that can effectively detect factual inconsistencies is crucial to reduce the propagation of misinformation and improve trust in model outputs. When testing on existing factual consistency benchmarks, we find that a few large language models (LLMs)... | With recent progress in generation capabilities of LLMs, automatic summarization is making its appearance in practical information consumption situations such as summarizing work meetings Prior work Recent investigations of using LLMs for evaluation have shown promising results across different NLP tasks To address thi... | Annotating Factuality of Summaries. With advances in language models and the increase in fluency and abstractiveness of summarizers, prior work showed that one of the key challenges in summarization was enforcing factual consistency Detecting Factual Errors. Some work has taken an automated approach to the detection of... | 1,134 | 1,457 | 1,134 |
Don't Mess with Mister-in-Between: Improved Negative Search for Knowledge Graph Completion | The best methods for knowledge graph completion use a 'dual-encoding' framework, a form of neural model with a bottleneck that facilitates fast approximate search over a vast collection of candidates. These approaches are trained using contrastive learning to differentiate between known positive examples and sampled ne... | A Knowledge Graph (KG) is a structured form of human knowledge consisting of entities, facts, relationships between any pair of entities, and semantic descriptions of entities. As important structures that store millions of data records that represent a part of human knowledge, KGs have been proven to bring substantial... | In this paper, we deal with the task of predicting missing entities in knowledge graph completion. Formally, given a knowledge graph G which has a set of entities E and predefined relations R, the tail entity retrieval task (h, r, ?) requires retrieving a list of entities {t 1 , t 2 , . . . , t k } from the entity set ... | 1,026 | 2,410 | 1,026 |
Transferable and Efficient: Unifying Dynamic Multi-Domain Product Categorization | As e-commerce platforms develop different business lines, a special but challenging product categorization scenario emerges, where there are multiple domain-specific category taxonomies and each of them evolves dynamically over time. In order to unify the categorization process and ensure efficiency, we propose a two-s... | Product categorization In real-world businesses, e-commerce platforms usually maintain multiple business lines with relatively independent taxonomies. These business lines are catering for different customer demands or specific domain applications, for example, one provides express delivery while another specializes in... | ∀i ∈ [1, n] domains, given a taxonomy G i with depth of d i and m leaf nodes, the path from root to leaf node forms the text which is regarded as hierarchical category label y (j) i (j ∈ [1, m]). For an input product title X i along with its meta concept labels {λ k }, our task is to output the correct category label i... | 1,026 | 2,453 | 1,026 |
PETALS: Collaborative Inference and Fine-tuning of Large Models | Many NLP tasks benefit from using large language models (LLMs) that often have more than 100 billion parameters. With the release of BLOOM-176B and OPT-175B, everyone can download pretrained models of this scale. Still, using these models requires high-end hardware unavailable to many researchers. In some cases, LLMs c... | In recent years, the NLP community has found that pretrained language models can solve many practical tasks, through either fine-tuning While the public availability of 100B+ parameter models makes them easier to access, they remain difficult to use for the majority of researchers and practitioners due to memory and co... | Practical usage of large language models can be broadly divided into two main scenarios: inference and parameter-efficient adaptation to downstream tasks. In this section, we outline the design of PETALS, showing how it handles both scenarios and also allows easily sharing trained adapters between the users of the syst... | 1,189 | 1,351 | 1,189 |
GammaE: Gamma Embeddings for Logical Queries on Knowledge Graphs | Embedding knowledge graphs (KGs) for multihop logical reasoning is a challenging problem due to massive and complicated structures in many KGs. Recently, many promising works projected entities and queries into a geometric space to efficiently find answers. However, it remains challenging to model the negation and unio... | Most important advances encode knowledge into large-scale graph data to model real-world knowledge graphs (KGs), such as Wikidata Knowledge graph reasoning can be represented by the first-order logic (FOL) queries with basic operators, such as the existential quantifier (∃), conjunction (∧), disjunction (∨), and negati... | This work is closely related to query embedding approaches Another method is to encode entities into the probabilistic density for performing multi-hop logical reasoning. A knowledge graph (KG) is a directed graph G = (V, E, R), where V is the set of entities, E is the set of triplets, and R denotes the set of relation... | 1,250 | 2,097 | 1,250 |
Multimodal Affective Analysis Using Hierarchical Attention Strategy with Word-Level Alignment | Multimodal affective computing, learning to recognize and interpret human affect and subjective information from multiple data sources, is still challenging because: (i) it is hard to extract informative features to represent human affects from heterogeneous inputs; (ii) current fusion strategies only fuse different mo... | With the recent rapid advancements in social media technology, affective computing is now a popular task in human-computer interaction. Sentiment analysis and emotion recognition, both of which require applying subjective human concepts for detection, can be treated as two affective computing subtasks on different leve... | Despite the large body of research on audio-visual affective analysis, there is relatively little work on combining text data. Early work combined human transcribed lexical features and low-level handcrafted acoustic features using feature-level fusion Our architecture is inspired by the document classification hierarc... | 784 | 1,700 | 784 |
SentiRec: Sentiment Diversity-aware Neural News Recommendation | Personalized news recommendation is important for online news services. Many news recommendation methods recommend news based on their relevance to users' historical browsed news, and the recommended news usually have similar sentiment with browsed news. However, if browsed news is dominated by certain kinds of sentime... | Online news websites such as Google news 1 have gained huge popularity for consuming digital news In this paper, we propose a sentiment diversityaware news recommendation approach named Sen-tiRec, which can improve the sentiment diversity of news recommendation by considering the sentiment orientation of candidate and ... | News recommendation is an important technique for online news websites to provide personalized news reading services In recent years, several news recommendation methods based on deep learning techniques are proposed In this section, we first present the formal definitions of the problem explored in this paper, then in... | 1,321 | 1,892 | 1,321 |
HaluEval: A Large-Scale Hallucination Evaluation Benchmark for Large Language Models | Large language models (LLMs), such as Chat-GPT, are prone to generate hallucinations, i.e., content that conflicts with the source or cannot be verified by the factual knowledge. To understand what types of content and to which extent LLMs are apt to hallucinate, we introduce the Hallucination Evaluation benchmark for ... | The advent of large language models (LLMs) Despite these prominent capabilities of LLMs trained on large text corpus, recent work has shown that LLMs are prone to suffer from hallucination generations across various applications To facilitate research in this direction, we present the Hallucination Evaluation benchmark... | Furthermore, for the task-specific examples, we design an automatic two-stage approach to generate hallucinated samples. First, based on existing task datasets (e.g., HotpotQA) as seed data, we employ ChatGPT to generate hallucinated samples with two styles of task-specific instructions, i.e., onepass and conversationa... | 1,134 | 708 | 1,134 |
SemFace: Pre-training Encoder and Decoder with a Semantic Interface for Neural Machine Translation | While pre-training techniques are working very well in natural language processing, how to pre-train a decoder and effectively leverage it for neural machine translation (NMT) still remains a tricky issue. The main reason is that the cross-attention module between the encoder and decoder cannot be pre-trained, and the ... | In recent years, pre-trained language models The above method essentially pre-trains a BERTlike In parallel to the idea of DALL•E Our contributions are listed as follows: • To the best of our knowledge, this is the first work to investigate and define a semantic interface between encoder and decoder for the MT pre-trai... | The overview of our proposed SemFace is illustrated in Figure The encoder is pre-trained to map the input from the monolingual semantic space into the interface, while the decoder is pre-trained to use the content from the interface via the cross attention module to finish decoding. The parameters of the encoder and th... | 1,295 | 746 | 1,295 |
Measuring and Mitigating Name Biases in Neural Machine Translation | Neural Machine Translation (NMT) systems exhibit problematic biases, such as stereotypical gender bias in the translation of occupation terms into languages with grammatical gender. In this paper we describe a new source of bias prevalent in NMT systems, relating to translations of sentences containing person names. To... | Natural language processing systems are seeing widespread adoption, prompting careful study into cultural biases they exhibit, and methods for bias mitigation. Gender bias is common in automated systems NMT systems are not only biased for gender, and gender bias is not limited to gender pronouns. Other biases include r... | • We show two new biases for person names in NMT, relating to gender and sentiment. In languages with rich grammatical gender, the gender of people referenced in a sentence will often affect the morphology of the other words in the sentence. For example, "[PER] is a Royal Designer" translates into German as either Masc... | 934 | 2,289 | 934 |
A Span-based Multimodal Variational Autoencoder for Semi-supervised Multimodal Named Entity Recognition | Multimodal named entity recognition (MNER) on social media is a challenging task which aims to extract named entities in free text and incorporate images to classify them into userdefined types. The existing semi-supervised named entity recognition methods focus on the text modal and are utilized to reduce labeling cos... | Multimodal named entity recognition (MNER) has become a fundamental task to extract named entities from unstructured texts and images on social media To reduce labeling costs in MNER, semisupervised learning is widely utilized to exploit the useful information of unlabeled data in text modal. Unlike the supervised sett... | For traditional named entity recognition, the labeled data is not always adequate because of the labeling costs. Therefore, semi-supervised learning is an important way to improve NER model performance without enough labeled data. Two widely used semi-supervised learning methods selftraining (ST) Considering to combine... | 1,360 | 2,207 | 1,360 |
A Review of Cross-Domain Text-to-SQL Models | WikiSQL and Spider, the large-scale crossdomain text-to-SQL datasets, have attracted much attention from the research community. The leaderboards of WikiSQL and Spider show that many researchers propose their models trying to solve the text-to-SQL problem. This paper first divides the top models in these two leaderboar... | Text-to-SQL is a task to translate the natural language query (input) written by users into the SQL query (output) automatically. For example, in Table 3, we want to input the question in the table into the model to get the SQL output. Early work on text-to-SQL focused on small-scale domainspecific databases such as Re... | We only discuss two paradigms achieving relatively high performance in the text-to-SQL task, shown in Figure 3.1 Paradigm One (Generate SQL structure ⇒ Fill schema) The most common text-to-SQL paradigm is to generate the SQL structure first and then fill the schema items (schema columns and tables). In WikiSQL, because... | 876 | 1,457 | 876 |
Agreement Prediction of Arguments in Cyber Argumentation for Detecting Stance Polarity and Intensity | In online debates, users express different levels of agreement/disagreement with one another's arguments and ideas. Often levels of agreement/disagreement are implicit in the text and must be predicted to analyze collective opinions. Existing stance detection methods predict the polarity of a post's stance toward a top... | Many major online and social media and networking sites, such as Facebook, Twitter, and Wikipedia, have taken over as the new public forum for people to discuss and debate issues of national and international importance. With more participants in these debates than ever before, the volume of unstructured discourse data... | Stance detection research has a wide interest in a variety of different application areas including opinion mining (Hasan and Ng, 2013), sentiment analysis For example, stance detection on Twitter often determines the author's stance (for/against/neutral) toward a proposition or target This dataset has many similaritie... | 1,406 | 4,186 | 1,406 |
Enhancing Authorship Attribution By Utilizing Syntax Tree Profiles | The aim of modern authorship attribution approaches is to analyze known authors and to assign authorships to previously unseen and unlabeled text documents based on various features. In this paper we present a novel feature to enhance current attribution methods by analyzing the grammar of authors. To extract the featu... | The increasing amount of documents available from sources like publicly available literary databases often raises the question of verifying disputed authorships or assigning authors to unlabeled text fragments. The original problem was initiated already in the midst of the twentieth century by Mosteller and Wallace, wh... | The basic idea of the approach is to utilize the syntax that is used by authors to distinguish authorships of text documents. Based on our previous work in the field of intrinsic plagiarism detection The number of choices an author has to formulate a sentence in terms of grammar is rather high, and the assumption in th... | 858 | 1,560 | 858 |
Back Transcription as a Method for Evaluating Robustness of Natural Language Understanding Models to Speech Recognition Errors | In a spoken dialogue system, an NLU model is preceded by a speech recognition system that can deteriorate the performance of natural language understanding. This paper proposes a method for investigating the impact of speech recognition errors on the performance of natural language understanding models. The proposed me... | Regardless of the near-human accuracy of automatic speech recognition in general-purpose transcription tasks, speech recognition errors can still significantly deteriorate the performance of a natural language understanding model that follows the speech-to-text module in a conversational system. The problem is even mor... | Data augmentation is a commonly employed method for improving the performance of neural models of vision, speech and language. Back translation The first experiments with augmenting ASR data with text-to-speech tools were conducted by Recently, there have been several papers addressing the issue of the robustness of na... | 689 | 2,096 | 689 |
How much complexity does an RNN architecture need to learn syntax-sensitive dependencies? | Long short-term memory (LSTM) networks and their variants are capable of encapsulating long-range dependencies, which is evident from their performance on a variety of linguistic tasks. On the other hand, simple recurrent networks (SRNs), which appear more biologically grounded in terms of synaptic connections, have ge... | For the last couple of decades, neural networks have been approached primarily from an engineering perspective, with the key motivation being efficiency, consequently moving further away from biological plausibility. Recent developments Recurrent Neural Networks (RNNs) have been used to analyze the principles and dynam... | There has been prior work on using LSTMs 2 Sentence taken from the dataset made available by From the biological point of view, According to Dale's principle, a neuron is either excitatory or inhibitory In the postsynaptic neuron, the integration of synaptic potentials is realized by the addition of excitatory (+ve) an... | 1,107 | 2,587 | 1,107 |
CLIReval: Evaluating Machine Translation as a Cross-Lingual Information Retrieval Task | We present CLIReval, an easy-to-use toolkit for evaluating machine translation (MT) with the proxy task of cross-lingual information retrieval (CLIR). Contrary to what the project name might suggest, CLIReval does not actually require any annotated CLIR dataset. Instead, it automatically transforms translations and ref... | Machine translation (MT) is the task of automatically translating sentences from a source language to a target language. A natural question that arises is how do we determine whether an MT system is translating sentences well? One answer is that we can engage human translators to evaluate the translated sentences manua... | Given a set of source documents S, an MT system φ converts S into a set of translated documents, T = φ(S) . Intrinsic MT metrics directly calculate an aggregated score between the sentences in T and sentences in R, where R is a set of reference documents. This approach makes several assumptions. First, CLIReval impleme... | 1,036 | 2,080 | 1,036 |
Language Independent Probabilistic Context-Free Parsing Bolstered by Machine Learning | Unlexicalized probabilistic context-free parsing is a general and flexible approach that sometimes reaches competitive results in multilingual dependency parsing even if a minimum of language-specific information is supplied. Furthermore, integrating parser results (good at long dependencies) and tagger results (good a... | Unlexicalized probabilistic context-free parsing is a simple and flexible approach that nevertheless has shown good performance | For development, we chose the initial sentences of every treebank, where is the number of the sentences in the test set. In this way, the sizes were realistic for the task. For parsing the test data, we added the development set to the training set. All the evaluations on the test sets were performed with the evaluatio... | 444 | 127 | 444 |
Is the Brain Mechanism for Hierarchical Structure Building Universal Across Languages? An fMRI Study of Chinese and English | Evidence from psycholinguistic studies suggests that the human brain builds a hierarchical syntactic structure during language comprehension. However, it is still unknown whether the neural basis of such structures is universal across languages. In this paper, we first analyze the differences in language structure betw... | A hallmark of human language ability is combining linear sequential word inputs into a hierarchical structure using abstract syntactic rules. This ability enables us to create infinite expressions from finite words. Previous studies have shown that several brain regions are involved in building the hierarchical syntact... | Building hierarchical syntactic structures is an important sub-process of language understanding. Existing work that has investigated this sub-process can be categorized into two groups. One is often called controlled experiments that design artificial stimuli to separate the brain activation, such as comparing structu... | 1,265 | 3,055 | 1,265 |
Continual Named Entity Recognition without Catastrophic Forgetting | Continual Named Entity Recognition (CNER) is a burgeoning area, which involves updating an existing model by incorporating new entity types sequentially. Nevertheless, continual learning approaches are often severely afflicted by catastrophic forgetting. This issue is intensified in CNER due to the consolidation of old... | Named Entity Recognition (NER) is a essential research area in Natural Language Understanding (NLU). Its purpose is to assign each token in a sequence with multiple entity types or non-entity type Deep learning approaches to CNER encounter two primary challenges. The first one is common to all continual learning method... | Continual Learning learns continuous tasks without reducing performance on previous tasks CNER Traditional NER focuses on the development of various deep learning models aimed at extracting entities from unstructured text CNER aims to train a model across t = 1, ..., T steps, progressively learning an expanding set of ... | 1,381 | 2,241 | 1,381 |
Logic-driven Indirect Supervision: An Application to Crisis Counseling | Ensuring the effectiveness of text-based crisis counseling requires observing ongoing conversations and providing feedback, both labor-intensive tasks. Automatic analysis of conversations-at the full chat and utterance levels-may help support counselors and provide better care. While some session-level training data (e... | Text-based crisis counseling services like Crisis Text Line Addressing the twin problems of managing counselor workload and ensuring quality requires training new counselors and providing feedback to existing ones. In particular, understanding suicide risk in client utterances may help counselors learn to prioritize hi... | In text-based crisis intervention, a client starts a chat session (also called an encounter) by typing a message, and the first available counselor replies to it. The session goes on till either the client finishes the conversation, or a certain amount of time elapses with no client response. The volume of messages to ... | 1,178 | 2,382 | 1,178 |
Facilitating Terminology Translation with Target Lemma Annotations | Most of the recent work on terminology integration in machine translation has assumed that terminology translations are given already inflected in forms that are suitable for the target language sentence. In day-to-day work of professional translators, however, it is seldom the case as translators work with bilingual g... | Translation into morphologically complex languages involves 1) making a lexical choice for a word in the target language and 2) finding its morphological form that is suitable for the morphosyntactic context of the target sentence. Most of the recent work on terminology translation, however, has assumed that the correc... | To train NMT systems that allow applying terminology constraints Our work is similar to work by Languages and Data. As our focus is on morphologically complex languages, in our experiments we translate from English into Latvian and Lithuanian (Baltic branch of the Indo-European language family) as well as Estonian (Fin... | 1,366 | 1,527 | 1,366 |
STACL: Simultaneous Translation with Implicit Anticipation and Controllable Latency using Prefix-to-Prefix Framework * | Simultaneous translation, which translates sentences before they are finished, is useful in many scenarios but is notoriously difficult due to word-order differences. While the conventional seq-to-seq framework is only suitable for full-sentence translation, we propose a novel prefix-to-prefix framework for simultaneou... | Simultaneous translation aims to automate simultaneous interpretation, which translates concurrently with the source-language speech, with a delay of only a few seconds. This additive latency is much more desirable than the multiplicative 2× slowdown in consecutive interpretation. With this appealing property, simultan... | As a very simple example within the prefix-toprefix framework, we present a wait-k policy, which first wait k source words, and then translates concurrently with the rest of source sentence, i.e., the output is always k words behind the input. This is inspired by human simultaneous interpreters who generally start tran... | 981 | 4,953 | 981 |
A Multi-Modal Context Reasoning Approach for Conditional Inference on Joint Textual and Visual Clues | Conditional inference on joint textual and visual clues is a multi-modal reasoning task that textual clues provide prior permutation or external knowledge, which are complementary with visual content and pivotal to deducing the correct option. Previous methods utilizing pretrained vision-language models (VLMs) have ach... | Cross modal reasoning is a hot research topic both in natural language processing and computer vision communities. Most cross modal reasoning tasks, such as Visual Question Answering Previous methods To verify the effectiveness of ModCR, we conduct extensive experiments on two cross modal reasoning data sets: PMR Our c... | Pretrained VLMs for Cross Modal Reasoning. Cross modal reasoning Vision-aided Language Models. Images can provide explicit and diverse visual information to improve the imaginative representation of language. Recent works show that vision-aided language models have achieved promising performance on natural language und... | 454 | 996 | 454 |
M2D2: A Massively Multi-Domain Language Modeling Dataset | We present M2D2, a fine-grained, massively multi-domain corpus for studying domain adaptation in language models (LMs). M2D2 consists of 8.5B tokens and spans 145 domains extracted from Wikipedia and Semantic Scholar. Using ontologies derived from Wikipedia and ArXiv categories, we organize the domains in each data sou... | Even though they can contain a wide variety of different types of domains, the texts that make up the corpora used to train and evaluate language models (LMs) are often treated as if they are all the same. This makes it challenging to characterize LM performance under diverse data distributions and understand how to ef... | Art Using M2D2, we investigate the following questions, as examples of the broad classes of new questions that can be asked: (1) how well do coarse and fine domains transfer to each other across the hierarchy? (2) which features and aspects of a domain are important for transfer? (3) how important is domain specificity... | 1,191 | 600 | 1,191 |
NeuInfer: Knowledge Inference on N-ary Facts | Knowledge inference on knowledge graph has attracted extensive attention, which aims to find out connotative valid facts in knowledge graph and is very helpful for improving the performance of many downstream applications. However, researchers have mainly poured attention to knowledge inference on binary facts. The stu... | With the introduction of connotative valid facts, knowledge inference on knowledge graph improves the performance of many downstream applications, such as vertical search and question answering In existing studies for knowledge inference on nary facts, each n-ary fact is represented as a group of peer attributes and at... | They can be divided into tensor/matrix based methods, translation based methods, and neural network based ones. The quintessential one of tensor/matrix based methods is RESCAL Translation based methods date back to TransE Neural network based methods model the validity of binary facts or the inference processes. For ex... | 1,067 | 2,584 | 1,067 |
Modeling Multi-hop Question Answering as Single Sequence Prediction | Fusion-in-decoder (FID) (Izacard and Grave, 2021) is a generative question answering (QA) model that leverages passage retrieval with a pre-trained transformer and pushed the state of the art on single-hop QA. However, the complexity of multi-hop QA hinders the effectiveness of the generative QA approach. In this work,... | Leveraging knowledge to make complex reasoning has been a fundamental problem of artificial intelligence. Open-domain question answering (QA) Recent work In this work, we propose PATHFID, a generative QA model that learns to generate an answer along with a reasoning path to improve its capability of multi-hop reasoning... | In this section, we formally introduce the problem setup and establish the necessary background. We first describe the multi-hop QA task in a general way. We assume that a collection of K passages are given for a question q: D q = {p 1 , p 2 , . . . , p K }, where D q can be a pre-defined set, or it can also be an outp... | 1,197 | 926 | 1,197 |
A Cluster-based Approach for Improving Isotropy in Contextual Embedding Space | The representation degeneration problem in Contextual Word Representations (CWRs) hurts the expressiveness of the embedding space by forming an anisotropic cone where even unrelated words have excessively positive correlations. Existing techniques for tackling this issue require a learning process to re-train models wi... | Despite their outstanding performance, CWRs are known to suffer from the so-called representation degeneration problem that makes the embedding space anisotropic To better understand the representation degeneration problem in pre-trained models, we analyzed the embedding space of GPT-2 In addition, we provide an analys... | Isotropy is a desirable property of word embedding spaces and arguably any other vector representation of data in general We measure the isotropy of embedding space using the partition function of where is a unit vector, is the corresponding embedding for the ℎ word in the embedding matrix W ∈ IR N×D , N is the number ... | 1,197 | 1,299 | 1,197 |
Prefix Lexicalization of Synchronous CFGs using Synchronous TAG | We show that an ε-free, chain-free synchronous context-free grammar (SCFG) can be converted into a weakly equivalent synchronous tree-adjoining grammar (STAG) which is prefix lexicalized. This transformation at most doubles the grammar's rank and cubes its size, but we show that in practice the size increase is only qu... | Greibach normal form (GNF; By using prefix lexicalized synchronous context-free grammars (SCFGs), This work investigates the formal properties of prefix lexicalized synchronous grammars as employed by | An SCFG is a tuple G = (N, Σ, P, S) where N is a finite nonterminal alphabet, Σ is a finite terminal alphabet, S ∈ N is a distinguished nonterminal called the start symbol, and P is a finite set of synchronous rules of the form (1) must be linked to exactly one nonterminal in α 2 , and vice versa. We write these links ... | 498 | 200 | 498 |
MultiEMO: An Attention-Based Correlation-Aware Multimodal Fusion Framework for Emotion Recognition in Conversations | Emotion Recognition in Conversations (ERC) is an increasingly popular task in the Natural Language Processing community, which seeks to achieve accurate emotion classifications of utterances expressed by speakers during a conversation. Most existing approaches focus on modeling speaker and contextual information based ... | Emotion Recognition in Conversations (ERC) is an emerging task in the field of Natural Language Processing (NLP), which aims to identify the emotion of each utterance in a conversation based on textual, audio and visual cues of the speaker. ERC has attracted an enormous amount of attention from both academia and indust... | Multimodal Fused Graph Convolutional Network (MMGCN) is proposed by The overall framework of MultiEMO is illustrated in Figure Existing research often adopts two different paradigms to extract contextualized textual features: (1) Two-stage paradigm To be specific, following Audio Feature Extraction: We follow Visual F... | 1,520 | 1,729 | 1,520 |
Hypergraph Transformer: Weakly-Supervised Multi-hop Reasoning for Knowledge-based Visual Question Answering | Knowledge-based visual question answering (QA) aims to answer a question which requires visually-grounded external knowledge beyond image content itself. Answering complex questions that require multi-hop reasoning under weak supervision is considered as a challenging problem since i) no supervision is given to the rea... | Visual question answering (VQA) is a semantic reasoning task that aims to answer questions about visual content depicted in images In this paper, we focus on the task which is called knowledge-based visual question answering, where a massive number of knowledge facts from a general knowledge base (KB) is given with an ... | Knowledge-based visual question answering Multi-hop knowledge graph reasoning is a process of sequential reasoning based on multiple evidences of a knowledge graph, and has been broadly used in various downstream tasks such as question answering To capture high-order semantics inherent in the knowledge sources, we ado... | 1,034 | 1,973 | 1,034 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.