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 Long Distance Slot Carryover in Spoken Dialogue Systems | Tracking the state of the conversation is a central component in task-oriented spoken dialogue systems. One such approach for tracking the dialogue state is slot carryover, where a model makes a binary decision if a slot from the context is relevant to the current turn. Previous work on the slot carryover task used mod... | In task-oriented spoken dialogue systems, the user and the system are engaged in interactions that can span multiple turns. A key challenge here is that the user can reference entities introduced in previous dialogue turns. For example, if a user request for what's the weather in arlington is followed by how about tomo... | A dialogue H is formulated as a sequence of utterances, alternatively uttered by a user (U) and the system agent (A): where each element h is an utterance. A subscript d denotes the utterance distance which measures the offset from the most recent user utterance (h U 0 ). The i-th token of an utterance with distance d ... | 1,002 | 839 | 1,002 |
CLIP Models are Few-shot Learners: Empirical Studies on VQA and Visual Entailment | CLIP has shown a remarkable zero-shot capability on a wide range of vision tasks. Previously, CLIP is only regarded as a powerful visual encoder. However, after being pretrained by language supervision from a large amount of image-caption pairs, CLIP itself should also have acquired some few-shot abilities for vision-l... | Vision-language understanding (VLU) tasks, such as visual question answering Recently, CLIP To answer the above question, in this work, we empirically study how to transfer CLIP's zero-shot ability into VLU tasks and further turn CLIP into a few-shot learner. We carried out experiments on two VLU tasks: 1) visual quest... | 2.1 CLIP CLIP, short for Contrastive Language-Image Pretraining T(text) • V(image), which is used as an alignment score between the input image and text. It is pretrained to distinguish aligned image-text pairs from randomly combined ones by a contrastive loss. Instead of training on vision benchmarks, CLIP leverages a... | 931 | 2,005 | 931 |
Learning Architectures from an Extended Search Space for Language Modeling | Neural architecture search (NAS) has advanced significantly in recent years but most NAS systems restrict search to learning architectures of a recurrent or convolutional cell. In this paper, we extend the search space of NAS. In particular, we present a general approach to learn both intra-cell and inter-cell architec... | Neural models have shown remarkable performance improvements in a wide range of natural language processing (NLP) tasks. Systems of this kind can broadly be characterized as following a neural network design: we model the problem via a pre-defined neural architecture, and the resulting network is treated as a black-box... | NAS is a promising method toward AutoML Despite of great success, previous studies restricted themselves to a small search space of neural networks. For example, most NAS systems were designed to find an architecture of recurrent or convolutional cell, but the remaining parts of the network are handcrafted In this work... | 973 | 2,731 | 973 |
LOCALRQA: From Generating Data to Locally Training, Testing, and Deploying Retrieval-Augmented QA Systems | Retrieval-augmented question-answering systems combine retrieval techniques with large language models to provide answers that are more accurate and informative. Many existing toolkits allow users to quickly build such systems using off-the-shelf models, but they fall short in supporting researchers and developers to c... | Retrieval-augmented question-answering (RQA) systems enhance large language models (LLMs) by enabling them to search through a large collection of documents before answering a user's query. These systems have shown improved performance in providing more accurate, informative, and factually grounded answers compared to ... | Haystack Table approaches RQA systems combine retrievers with powerful LLMs to provide answers that are more accurate and informative. Given a user query, a retriever first selects k most relevant passages from a collection of documents. Then, a generative model produces an answer conditioned on the user's query, selec... | 813 | 1,079 | 813 |
A* shortest string decoding for non-idempotent semirings | The single shortest path algorithm is undefined for weighted finite-state automata over nonidempotent semirings because such semirings do not guarantee the existence of a shortest path. However, in non-idempotent semirings admitting an order satisfying a monotonicity condition (such as the plus-times or log semirings),... | Weighted finite-state automata provide a compact representation of hypotheses in various speech recognition and text processing applications (e.g., The shortest path-and the algorithms that compute it-are well-defined when the weights of a lattice are idempotent and exhibit the path property. These properties are forma... | Before we introduce the proposed decoding algorithm we provide definitions of key notions. Weighted automata algorithms operate with respect to an algebraic system known as a semiring, characterized by the combination of two monoids. Definition 2.1. A monoid is a pair (K, •) where K is a set and • is a binary operator ... | 875 | 949 | 875 |
Polar Ducks and Where to Find Them: Enhancing Entity Linking with Duck Typing and Polar Box Embeddings | Entity linking methods based on dense retrieval are widely adopted in large-scale applications for their efficiency, but they can fall short of generative models, as they are sensitive to the structure of the embedding space. To address this issue, this paper introduces DUCK, an approach to infusing structural informat... | State-of-the-art approaches to entity linking, namely the task of linking mentions of entities in a text to the corresponding entries in a knowledge base (KB) In this paper, we aim to close the gap with generative approaches by infusing structural information in the latent space of retrieval-based methods. Recent work ... | The Italy national football team has represented Italy in football since 1910. Italy, officially the Italian Republic or the Republic of Italy, is a country in Europe. We achieve this goal by drawing inspiration from the concept of duck typing in programming languages, which relies on the idea of defining the type of a... | 1,122 | 426 | 1,122 |
Learning to Control the Fine-grained Sentiment for Story Ending Generation | Automatic story ending generation is an interesting and challenging task in natural language generation. Previous studies are mainly limited to generate coherent, reasonable and diversified story endings, and few works focus on controlling the sentiment of story endings. This paper focuses on generating a story ending ... | Story ending generation aims at completing the plot and concluding a story given a story context. Previous works mainly study on how to generate a coherent, reasonable and diversified story ending | She almost won the game, but eventually lost. The game ended with a draw. She eventually won the game. She won the game and was very proud of her team. Story context: Sally really loves to play soccer. She joined a team with her friends and she plays everyday. Her coach and her teammates are all really fun. Sally pract... | 1,371 | 196 | 1,371 |
Shapley Head Pruning: Identifying and Removing Interference in Multilingual Transformers | Multilingual transformer-based models demonstrate remarkable zero and few-shot transfer across languages by learning and reusing language-agnostic features. However, as a fixed-size model acquires more languages, its performance across all languages degrades. Those who attribute this interference phenomenon to limited ... | Cross-lingual transfer learning aims to utilize a natural language processing system trained on a source language to improve results for the same task in a different target language. The core goal is to maintain relevant learned patterns from the source while disregarding those which are inapplicable to the target. Mul... | In a qualitative study, we find that the most language-agnostic heads identified have a visible language-agnostic function, while language differences can be measured meaningfully for language-specific heads. 2 Related Work A large amount of work has studied both the theoretical underpinnings of learning common structu... | 1,016 | 1,734 | 1,016 |
R2D2: Recursive Transformer based on Differentiable Tree for Interpretable Hierarchical Language Modeling | Human language understanding operates at multiple levels of granularity (e.g., words, phrases, and sentences) with increasing levels of abstraction that can be hierarchically combined. However, existing deep models with stacked layers do not explicitly model any sort of hierarchical process. This paper proposes a recur... | The idea of devising a structural model of language capable of learning both representations and meaningful syntactic structure without any humanannotated trees has been a long-standing but challenging goal. Across a diverse range of linguistic theories, human language is assumed to possess a recursive hierarchical str... | Here, k is a split point from i to j -1, f (•) is a composition function that we shall further define later on, p k i,j and p k i,j denote the single step combination probability and the subtree probability, respectively, at split point k, p i,j and p i,j are the concatenation of all p k i,j or p k i,j values, and GUMB... | 874 | 1,425 | 874 |
Dependency resolution at the syntax-semantics interface: psycholinguistic and computational insights on control dependencies | Using psycholinguistic and computational experiments we compare the ability of humans and several pre-trained masked language models to correctly identify control dependencies in Spanish sentences such as 'José le prometió/ordenó a María ser ordenado/a' ('Joseph promised/ordered Mary to be tidy'). These structures unde... | Treating pre-trained language models (LMs) as psycholinguistic subjects via the behavioral evaluation of their probability distributions has proven to be a very useful strategy to study to which extent they are able to generalize grammatical information from raw text (1) a. María i f le prometió a José j m ser ordenada... | Targeted evaluation of LMs: Targeted evaluations of LMs focusing on different syntactic phenomena have found evidence suggesting that these models may generalize syntactic information from raw text Despite the fact that most of the work evaluating the linguistic capabilities of LMs has been carried out in English, ther... | 947 | 2,933 | 947 |
CLAD-ST: Contrastive Learning with Adversarial Data for Robust Speech Translation | The cascaded approach continues to be the most popular choice for speech translation (ST). This approach consists of an automatic speech recognition (ASR) model and a machine translation (MT) model that are used in a pipeline to translate speech in one language to text in another language. MT models are often trained o... | Neural machine translation (NMT) has made significant advancements over the past several years with claims of achieving 'human parity' Robustness is especially important in cascaded speech translation (ST) systems, where an NMT model works on the output of the upstream automatic speech recognition (ASR) system. In this... | Our NMT model is a Transformer model To improve the robustness of the NMT model on noisy ASR outputs for cascaded speech translation, we use a contrastive learning method To get the encoder sentence representations efficiently for contrastive learning, a [CLS] token is prepended to the input sentences similar to the BE... | 1,216 | 1,700 | 1,216 |
Automatic Construction of Machine Translation Knowledge Using Translation Literalness | When machine translation (MT) knowledge is automatically constructed from bilingual corpora, redundant rules are acquired due to translation variety. These rules increase ambiguity or cause incorrect MT results. To overcome this problem, we constrain the sentences used for knowledge extraction to "the appropriate bilin... | Along with the efforts made to accumulate bilingual corpora for many language pairs, quite a few machine translation (MT) systems that automatically construct their knowledge from corpora have been proposed Such rules increase ambiguity and may cause inappropriate MT results. Translation variety increases with corpus s... | First, we describe the problems inherent in bilingual corpora when we automatically construct MT knowledge. Some bilingual sentences in corpora depend on the context or situation, and these are not always correct in different contexts. For instance, the English determiner 'the' is not generally translated into Japanese... | 537 | 1,376 | 537 |
LSTMs Can Learn Syntax-Sensitive Dependencies Well, But Modeling Structure Makes Them Better | Language exhibits hierarchical structure, but recent work using a subject-verb agreement diagnostic argued that state-ofthe-art language models, LSTMs, fail to learn long-range syntax-sensitive dependencies. Using the same diagnostic, we show that, in fact, LSTMs do succeed in learning such dependencies-provided they h... | Recurrent neural networks (RNNs) are remarkably effective models of sequential data. Recent years have witnessed the widespread adoption of recurrent architectures such as LSTMs Here we revisit the question asked by Contrary to the findings of Given the strong performance of word-based LSTM language models, are there a... | We revisit the number agreement task with LSTMs trained on language modeling objectives, as proposed by While the pretrained large-scale language model of Prior work has confirmed the notion that, in many cases, statistical models are able to achieve good performance under some aggregate metric by overfitting to patter... | 962 | 1,663 | 962 |
Text and Causal Inference: A Review of Using Text to Remove Confounding from Causal Estimates | Many applications of computational social science aim to infer causal conclusions from nonexperimental data. Such observational data often contains confounders, variables that influence both potential causes and potential effects. Unmeasured or latent confounders can bias causal estimates, and this has motivated intere... | In contrast to descriptive or predictive tasks, causal inference aims to understand how intervening on one variable affects another variable strongly biased estimates and thus invalid causal conclusions. To eliminate confounding bias, one approach is to perform randomized controlled trials (RCTs) in which researchers r... | In Table Text as a surrogate for confounders. Traditionally, causal research that uses human subjects as the unit of analysis would infer demographics via surveys. However, with the proliferation of the web and social media, social research now includes large-scale observational data that would be challenging to obtain... | 935 | 2,993 | 935 |
Copyright Violations and Large Language Models | Language models may memorize more than just facts, including entire chunks of texts seen during training. Fair use exemptions to copyright laws typically allow for limited use of copyrighted material without permission from the copyright holder, but typically for extraction of information from copyrighted materials, ra... | If you remember what Pride and Prejudice is about, you have not necessarily memorized it. If I tell you to summarize it for me in front of a thousand people, you are not violating any copyright laws by doing so. If you write it down for me, word by word, handing out copies to everyone in the room, it would be a differe... | The trade-off between memorization and generalization Based on how memorization is distributed, and what is predictive thereof, Copyright laws and conventions grant the creators of a work exclusive rights to use and distribute their creations, with certain exceptions In a European context, quotation is listed as one of... | 960 | 2,138 | 960 |
Keywords and Instances: A Hierarchical Contrastive Learning Framework Unifying Hybrid Granularities for Text Generation | Contrastive learning has achieved impressive success in generation tasks to militate the "exposure bias" problem and discriminatively exploit the different quality of references. Existing works mostly focus on contrastive learning on the instance-level without discriminating the contribution of each word, while keyword... | Generation tasks such as storytelling, paraphrasing, and dialogue generation aim at learning a certain correlation between text pairs that maps an arbitrary-length input to another arbitrary-length output. Traditional methods are mostly trained with "teacher forcing" and lead to an "exposure bias" problem Based on the ... | Contrastive learning is used to learn representations by teaching the model which data points are similar or not. Due to the excellent performance on self-supervised and semi-supervised learning, it has been widely used in natural language processing (NLP). Firstly, The Mahalanobis distance is a measure of the distance... | 1,242 | 1,532 | 1,242 |
Using Neural Machine Translation Methods for Sign Language Translation | We examine methods and techniques, proven to be helpful for the text-to-text translation of spoken languages in the context of gloss-to-text translation systems, where the glosses are the written representation of the signs. We present one of the first works that include experiments on both parallel corpora of the Germ... | Sign languages (SL), the main medium of exchanging information for the deaf and the hard of hearing, are visual-spatial natural languages with their own linguistic rules. In contrast to the spoken ones, they lack a written form, on one hand, and use face, hands and body to convey meaning, on the other. However, in our ... | Sign language translation is a relatively new research field with recent findings made possible thanks to the continuous advances in neural machine translation (NMT). Several experiments with SL gloss-to-text translation have taken place in the previous decade using statistical phrase-based machine translation To the b... | 915 | 1,916 | 915 |
Can Pretrained Language Models (Yet) Reason Deductively? | Acquiring factual knowledge with Pretrained Language Models (PLMs) has attracted increasing attention, showing promising performance in many knowledge-intensive tasks. Their good performance has led the community to believe that the models do possess a modicum of reasoning competence rather than merely memorising the k... | Pretrained Language Models (PLMs) such as BERT (2) (3) (1) (2) (3) (1) (2) (3) Automatic reasoning, a systematic process of deriving previously unknown conclusions from given formal representations of knowledge In particular, deductive reasoning 2 is one of the most promising directions Despite promising applications o... | Knowledge Probing, Infusing, and Editing with PLMs. PLMs appear to memorise (world) knowledge facts during pretraining, and such captured knowledge is useful for knowledge-intensive tasks Knowledge Reasoning with PLMs. In recent years, PLMs have also achieved impressive progress in knowledge reasoning Although some res... | 1,208 | 1,311 | 1,208 |
A Supertag-Context Model for Weakly-Supervised CCG Parser Learning | Combinatory Categorial Grammar (CCG) is a lexicalized grammar formalism in which words are associated with categories that specify the syntactic configurations in which they may occur. We present a novel parsing model with the capacity to capture the associative adjacent-category relationships intrinsic to CCG by param... | Learning parsers from incomplete or indirect supervision is an important component of moving NLP research toward new domains and languages. But with less information, it becomes necessary to devise ways of making better use of the information that is available. In general, this means constructing inductive biases that ... | In the CCG formalism, every constituent, including those at the lexical level, is associated with a structured CCG category that defines that constituent's relationships to the other constituents in the sentence. Categories are defined by a recursive structure, where a category is either atomic (possibly with features)... | 845 | 3,263 | 845 |
Information Presentation in Spoken Dialogue Systems | To tackle the problem of presenting a large number of options in spoken dialogue systems, we identify compelling options based on a model of user preferences, and present tradeoffs between alternative options explicitly. Multiple attractive options are structured such that the user can gradually refine her request to f... | The goal of spoken dialogue systems (SDS) is to offer efficient and natural access to applications and services, such as email and calendars, travel and entertainment booking, and product recommendation. In evaluating nine SDS in the DARPA Communicator domain (flight, hotel, and rental car hire), A return flight is wit... | Previous work in natural language generation showed how a multi-attribute decision-theoretic model of user preferences could be used to determine the attributes that are most relevant to mention when generating recommendations tailored to a particular user However, there are several limitations to this approach. First,... | 695 | 2,506 | 695 |
Compounding Geometric Operations for Knowledge Graph Completion | Geometric transformations including translation, rotation, and scaling are commonly used operations in image processing. Besides, some of them are successfully used in developing effective knowledge graph embedding (KGE). Inspired by the synergy, we propose a new KGE model by leveraging all three operations in this wor... | Knowledge graphs (KGs) such as DBpedia Geometric operations such as translation and rotation belong to the family of affine transformations. These operations have been used to build effective KGE models such as TransE, RotatE, and PairRE. Previous KGEs often use a single type of operation to model all relation patterns... | Distance-based scoring function is a prevailing strategy in optimizing KGE. The main idea is to model a relation as a transformation that places head entity vectors in the proximity of their corresponding tail entity vectors, and vice versa. For a given triple, (h, r, t), the goal is to minimize the distance between h ... | 1,007 | 1,967 | 1,007 |
Semantic Simplification for Sentiment Classification | Recent work on document-level sentiment classification has shown that the sentiment in the original text is often hard to capture, since the sentiment is usually either expressed implicitly or shifted due to the occurrences of negation and rhetorical words. To this end, we enhance the original text with a sentiment-dri... | As a critical application of natural language processing, document-level sentiment classification has received considerable attention during the last two decades with the underlying assumption that the entire text has an overall polarity. In the literature, previous studies focus on predicting the overall sentiment fro... | In this study, we introduce two related topics of this study: document-level sentiment classification and text simplification. Finally, we employ the pre-trained language model BERT where X is the original text, Y is the generated simplified clause, [CLS] is BERT's special classification token, and [SEP ] is the specia... | 898 | 2,077 | 898 |
Achieving Conversational Goals with Unsupervised Post-hoc Knowledge Injection | A limitation of current neural dialog models is that they tend to suffer from a lack of specificity and informativeness in generated responses, primarily due to dependence on training data that covers a limited variety of scenarios and conveys limited knowledge. One way to alleviate this issue is to extract relevant kn... | Generic responses which lack specificity have been a major issue in existing dialog models We propose and evaluate an approach for unsupervised knowledge injection into a dialog model's response at decoding time We experiment with two types of knowledge sources: language models, which we treat as parametric knowledge b... | Our goal is to construct a dialog response by injecting knowledge (from external textual sources) at decoding time, without having to retrain the models. Consider a dialog model M from which we can sample a dialog response x d given a dialog history H. We shall refer to the response x d sampled from such a model withou... | 1,225 | 1,477 | 1,225 |
Multilingual Speech Translation from Efficient Finetuning of Pretrained Models | We present a simple yet effective approach to build multilingual speech-to-text (ST) translation through efficient transfer learning from a pretrained speech encoder and text decoder. Our key finding is that a minimalistic LNA (LayerNorm and Attention) finetuning can achieve zero-shot crosslingual and crossmodality tra... | Recent advances in pretraining over unlabeled data and then finetuning on labeled data leads to significant performance improvement in text understanding and generation tasks Our contributions are as follows: • We propose a simple and effective approach to combine pretrained single-modality modules to perform speech-to... | Our model leverages a pretrained wav2vec 2.0 We add a lightweight adaptor module in between encoder and decoder to better align the two mod-ules pretrained with different modalities. The adaptor module performs projection and downsampling to alleviate length inconsistency between the audio and text sequences. Specifica... | 1,069 | 1,442 | 1,069 |
Leap-of-Thought: Accelerating Transformers via Dynamic Token Routing | Computational inefficiency in transformers has been a long-standing challenge, hindering the deployment in resource-constrained or realtime applications. One promising approach to mitigate this limitation is to progressively remove less significant tokens, given that the sequence length strongly contributes to the inef... | The advent of Transformer One typical approach to tackle this challenge is to reduce the number of tokens processed within transformer layers In this paper, we propose Leap-of-Thought (LoT) LoT offers several advantages compared to the permanent removal. Primarily, LoT has the potential to mitigate the risk of losing c... | In this section, we mainly review the methods that adaptively control the computation in pre-trained language models. Recent approaches can be classified into two categories: width-wise and depthwise approaches. The former focuses on reducing the number of tokens processed by transformers, while the latter aims to decr... | 1,154 | 1,681 | 1,154 |
LM-CPPF: Paraphrasing-Guided Data Augmentation for Contrastive Prompt-Based Few-Shot Fine-Tuning | In recent years, there has been significant progress in developing pre-trained language models for NLP. However, these models often struggle when fine-tuned on small datasets. To address this issue, researchers have proposed various adaptation approaches. Promptbased tuning is arguably the most common way, especially f... | Pre-trained language models (PLMs) are trained on large-scaled corpora in a self-supervised fashion. They have fundamentally changed the NLP community in the past few years by achieving impressive results in various Tasks By the introduction of GPT-3 Prompt-based fine-tuning is a method for adapting PLMs to specific ta... | LLMs like GPT-3 Paraphrasing is the task of expressing the same meaning with different words or structures. It can be used to create training data with increased diversity and naturalness for NLP tasks, such as text classification Background Contrastive learning's success relies on data augmentation, which creates new ... | 1,277 | 793 | 1,277 |
Deep Neural Model Inspection and Comparison via Functional Neuron Pathways | We introduce a general method for the interpretation and comparison of neural models. The method is used to factor a complex neural model into its functional components, which are comprised of sets of co-firing neurons that cut across layers of the network architecture, and which we call neural pathways. The function o... | Interpretation of neural models is a difficult task because the knowledge learned within neural networks is distributed across hundreds of thousands of parameters. Interpreting the significance of any individual neuron is tantamount to reconstructing a forest based on a single pine needle. More specifically, the contri... | Our work falls under the broad topic of neural network interpretation. Recently, in this area of research a wide variety of models have been the target of investigation, including additive classifiers We observe that neural interpretation approaches fall within several broad categories: visualizations and heatmaps Rece... | 768 | 1,746 | 768 |
Enhancing Word Embeddings with Knowledge Extracted from Lexical Resources | In this work, we present an effective method for semantic specialization of word vector representations. To this end, we use traditional word embeddings and apply specialization methods to better capture semantic relations between words. In our approach, we leverage external knowledge from rich lexical resources such a... | Vector representations of words (embeddings) have become the cornerstone of modern Natural Language Processing (NLP), as learning word vectors and utilizing them as features in downstream NLP tasks is the de facto standard. Word embeddings To summarize, our contributions in this work are as follows: • We introduce a se... | Numerous methods have been introduced for incorporating structured linguistic knowledge from external resources to word embeddings. Fundamentally, there exist three categories of semantic specialization approaches: (a) joint methods which incorporate lexical information during the training of distributional word vector... | 573 | 795 | 573 |
Penn & BGU BabyBERTa+ for Strict-Small BabyLM Challenge | The BabyLM Challenge aims at pre-training a language model on a small-scale dataset of inputs intended for children. In this work, we adapted the architecture and masking policy of BabyBERTa (Huebner et al., 2021) to solve the strict-small track of the BabyLM challenge. Our model, Penn & BGU BabyBERTa+, was pre-traine... | With the emergence of deep-learning techniques | In this section, we provide the descriptions of our BabyBERTa+ model including the architectures, tokenizers, training objectives and so on. As shown in Table Following previous work To train the masked language model, the standard RoBERTa masking strategy replaces 80% of the corrupted tokens with the "<mask>" token, w... | 547 | 46 | 547 |
ArgAnalysis35K : A large-scale dataset for Argument Quality Analysis | Argument Quality Detection is an emerging field in NLP which has seen significant recent development. However, existing datasets in this field suffer from a lack of quality, quantity and diversity of topics and arguments, specifically the presence of vague arguments that are not persuasive in nature. In this paper, we ... | Parliamentary Debate is an extemporaneous form of debating. One of the major intersections of Natural Language Processing and Debating was IBM Project Debater The dimension that we introduce here is a detailed explanation of why the argument made is true, applicable or impactful, henceforth referred to as "analysis". A... | We should end racial profiling Argument relevance is an important indicator of persuasiveness according to Application of Instance-based annotator reliability to argumentation is another important contribution described in this paper. Some annotators might know a lot more about art than about the criminal justice syste... | 890 | 2,401 | 890 |
HIT-SCIR at MRP 2019: A Unified Pipeline for Meaning Representation Parsing via Efficient Training and Effective Encoding | This paper describes our system (HIT-SCIR) for the CoNLL 2019 shared task: Cross-Framework Meaning Representation Parsing. We extended the basic transition-based parser with two improvements: a) Efficient Training by realizing stack LSTM parallel training; b) Effective Encoding via adopting deep contextualized word emb... | The goal of the CoNLL 2019 shared task Recently, a lot of semantic graphbanks arise, which differ in the design of graphs Most semantic parsers are only designed for one or few specific graphbanks, due to the differences in annotation schemes. For example, the currently best parser for SDP is graph-based Therefore, the... | Our system architecture is shown in Figure In order to design the unified transition-based parser, we refer to the following frameworkspecific parsers: A tuple (S, L, B, E, V ) is used to represent parsing state, where S is a stack holding processed words, L is a list holding words popped out of S that will be pushed b... | 680 | 1,332 | 680 |
Information-Theoretic Probing for Linguistic Structure | The success of neural networks on a diverse set of NLP tasks has led researchers to question how much these networks actually "know" about natural language. Probes are a natural way of assessing this. When probing, a researcher chooses a linguistic task and trains a supervised model to predict annotations in that lingu... | Neural networks are the backbone of modern stateof-the-art natural language processing (NLP) systems. One inherent by-product of training a neural network is the production of real-valued representations. Many speculate that these representations encode a continuous analogue of discrete linguistic properties, e.g., par... | Following Let S be a random variable ranging over all possible sequences of words. For the sake of this paper, we assume the vocabulary V is finite and, thus, the values S can take are in V * . We write s ∈ S as s = s 1 • • • s |s| for a specific sentence, where each s i ∈ V is a specific token in the sentence at the p... | 1,375 | 1,376 | 1,375 |
ELLEIPO: A module that computes coordinative ellipsis for language generators that don't | Many current sentence generators lack the ability to compute elliptical versions of coordinated clauses in accordance with the rules for Gapping, Forward and Backward Conjunction Reduction, and SGF (Subject Gap in clauses with Finite/Fronted verb). We describe a module (implemented in JAVA, with German and Dutch as tar... | Coordination and coordinative ellipsis are essential tools for the sentence aggregation component of any language generator. Very often, when the aggregator chooses to combine several clauses into a single coordinate structure, the need arises to eliminate unnatural reduplications of coreferential constituents. In the ... | ELLEIPO is loosely based on Kempen's (subm.) psycholinguistically motivated syntactic theory of clausal coordination and coordinative ellipsis. It departs from the assumption that the generator's strategic (conceptual, pragmatic) component is responsible for selecting the concepts and conceptual structures that enable ... | 825 | 1,576 | 825 |
Re-evaluating the Role of BLEU in Machine Translation Research | We argue that the machine translation community is overly reliant on the Bleu machine translation evaluation metric. We show that an improved Bleu score is neither necessary nor sufficient for achieving an actual improvement in translation quality, and give two significant counterexamples to Bleu's correlation with hum... | Over the past five years progress in machine translation, and to a lesser extent progress in natural language generation tasks such as summarization, has been driven by optimizing against n-grambased evaluation metrics such as Bleu However, there is a question as to whether minimizing the error rate with respect to Ble... | The rationale behind the development of Bleu The way that Bleu and other automatic evaluation metrics work is to compare the output of a machine translation system against reference human translations. Machine translation evaluation metrics differ from other metrics that use a reference, like the word error rate metric... | 468 | 1,707 | 468 |
Learning Robust Representations for Continual Relation Extraction via Adversarial Class Augmentation | Continual relation extraction (CRE) aims to continually learn new relations from a classincremental data stream. CRE model usually suffers from catastrophic forgetting problem, i.e., the performance of old relations seriously degrades when the model learns new relations. Most previous work attributes catastrophic forge... | Relation extraction (RE) aims to detect the relation of two given entities in a sentence. Traditional RE models are trained on a fixed dataset with a predefined relation set, which cannot handle the reallife situation where new relations are constantly emerging. To this end, continual relation extraction (CRE) the mode... | Relation Extraction Conventional Relation Extraction (RE) focuses on extracting the predefined relation of two given entities in a sentence. Recently, a variety of deep neural networks (DNN) have been proposed for RE, mainly including: 1) Convolutional or Recurrent neural network (CNN or RNN) based methods Continual Le... | 1,150 | 1,928 | 1,150 |
Predicting and Using Target Length in Neural Machine Translation | Attention-based encoder-decoder models have achieved great success in neural machine translation tasks. However, the lengths of the target sequences are not explicitly predicted in these models. This work proposes length prediction as an auxiliary task and set up a sub-network to obtain the length information from the ... | In recent years, neural network (NN) models have achieved great improvements in machine translation (MT) tasks. Despite the success achieved in neural machine translation (NMT), current NMT systems do not model the length of the output explicitly, and thus various length normalization approaches are often used in decod... | Multi-task learning is an important training strategy that aims to improve the generalization performance of the main task with some other related tasks To predict the target length based on the standard transformer architecture We predict the length of the target sequence by a classifier in the range of We also embed ... | 546 | 1,852 | 546 |
Lexical Morphology in Machine Translation: a Feasibility Study | This paper presents a feasibility study for implementing lexical morphology principles in a machine translation system in order to solve unknown words. Multilingual symbolic treatment of word-formation is seducing but requires an in-depth analysis of every step that has to be performed. The construction of a prototype ... | Formalising morphological information to deal with morphologically constructed unknown words in machine translation seems attractive, but raises many questions about the resources and the prerequisites (both theoretical and practical) that would make such symbolic treatment efficient and feasible. In this paper, we des... | Unknown words are a problematic issue in any NLP tool. Depending on the studies Usually, three main groups of unknown words are distinguished: proper names, errors, and neologisms, and the possible solution highly depends on the type of unknown word to be solved. In this paper, we concentrate on neologisms which are co... | 513 | 1,404 | 513 |
DIONYSUS: A Pre-trained Model for Low-Resource Dialogue Summarization | Dialogue summarization has recently garnered significant attention due to its wide range of applications. However, existing methods for summarizing dialogues have limitations because they do not take into account the inherent structure of dialogue and rely heavily on labeled data, which can lead to poor performance in ... | Text summarization aims to produce concise and accurate summaries of long texts. Recent research on pre-trained neural language models has shown success in summarizing monologues Self-supervised text summarization models To address these challenges, we propose DIONY-SUS, a pre-trained sequence-to-sequence model designe... | Figure In certain types of dialogue, such as daily chats, it can be challenging to gather all necessary information from just a few dialogue turns due to the dispersed nature of dialogue information. To address this problem, we have created a summary helper model that generates pseudo summaries for each training exampl... | 1,104 | 2,551 | 1,104 |
Leveraging Passage Retrieval with Generative Models for Open Domain Question Answering | Generative models for open domain question answering have proven to be competitive, without resorting to external knowledge. While promising, this approach requires to use models with billions of parameters, which are expensive to train and query. In this paper, we investigate how much these models can benefit from ret... | Recently, several works have shown that factual information can be extracted from large scale language models trained on vast quantities of data Retrieval based approaches were previously considered in the context of open domain question answering with extractive models support documents, before extracting the answer f... | Open domain question answering is the task of answering general domain questions, in which the evidence is not given as input to the system. While being a longstanding problem in natural language processing Passage retrieval is an important step in open domain question answering, and is an active area of research to im... | 748 | 1,479 | 748 |
"The Boating Store Had Its Best Sail Ever": Pronunciation-attentive Contextualized Pun Recognition | Humor plays an important role in human languages and it is essential to model humor when building intelligence systems. Among different forms of humor, puns perform wordplay for humorous effects by employing words with double entendre and high phonetic similarity. However, identifying and modeling puns are challenging ... | During the last decades, social media has promoted the creation of a vast amount of humorous web contents These two forms of puns have been studied in literature from different angles. To recognize puns in a sentence, word sense disambiguation techniques (WSD) In this paper, we propose Pronunciation-attentive Contextua... | Pun Recognition and Generation To recognize puns, In this section, we first formally define the problem and then introduce the proposed method, PCPR. Suppose the input text consists of a sequence of N words {w 1 , w 2 , • • • , w N }. For each word w i with M i phonemes in its pronunciation, the phonemes are denoted as... | 1,008 | 1,256 | 1,008 |
A Relational Memory-based Embedding Model for Triple Classification and Search Personalization | Knowledge graph embedding methods often suffer from a limitation of memorizing valid triples to predict new ones for triple classification and search personalization problems. To this end, we introduce a novel embedding model, named R-MeN, that explores a relational memory network to encode potential dependencies in re... | Knowledge graphs (KGs) -representing the genuine relationships among entities in the form of triples (subject, relation, object) denoted as (s, r, o) -are often insufficient for knowledge presentation due to the lack of many valid triples Early embedding models such as TransE Existing embedding models are showing promi... | Table Regarding TransE, we obtain the second-best accuracy of 89.2% on WN11 and a competitive accuracy of 88.1% on FB13. Figure Method WN11 FB13 Avg. NTN Table Method MRR H@1 SE (Original rank) 0.559 38.5 CI Table Next, we present in Figure For the last experiment, we compute and report our ablation results over 2 fa... | 943 | 3,244 | 943 |
Weakly-Supervised Questions for Zero-Shot Relation Extraction | Zero-Shot Relation Extraction (ZRE) is the task of Relation Extraction where the training and test sets have no shared relation types. This very challenging domain is a good test of a model's ability to generalize. Previous approaches to ZRE reframed relation extraction as Question Answering (QA), allowing for the use ... | Building models that capture abstract knowledge rather than just memorizing data is one of the inspirations for zero-shot benchmarks In ZRE, the test relations do not appear in the training data, so one cannot apply typical relation classification approaches. One method for ZRE is to reframe the task as a Question-Answ... | In this work, we train models to extract facts from unstructured text. Facts are represented as triplets (e 1 , r, e 2 ) where e 1 is the head entity, e 2 is the tail entity, and r is the relation keywords. We explore relation extraction To transfer the models pre-trained on QA corpora to the task of TE, we can provide... | 1,285 | 1,337 | 1,285 |
Self-Adaptive Scaling Approach for Learnable Residual Structure | Residual has been widely applied to build deep neural networks with enhanced feature propagation and improved accuracy. In the literature, multiple variants of residual structure are proposed. However, most of them are manually designed for particular tasks and datasets and the combination of existing residual structur... | Recently, residual learning attracts considerable attention in training deep neural networks, and many efforts have been devoted to study the utilization of residual structure in tasks across a broad span of fields, including but not limited to computer vision Generally, the residual structures (as illustrated in Figur... | In recent years, the application of residual structure to deep neural networks has become an active research topic The first is how should the information from the skip connection and the residual branch be well balanced so that the best improvements can be achieved. The second is how should the neural network with res... | 1,311 | 3,119 | 1,311 |
Elastic weight consolidation for better bias inoculation | The biases present in training datasets have been shown to affect models for sentence pair classification tasks such as natural language inference (NLI) and fact verification. While fine-tuning models on additional data has been used to mitigate them, a common issue is that of catastrophic forgetting of the original tr... | A number of recent works have illustrated shortcomings in sentence-pair classification models that are used for Natural Language Inference (NLI). These arise from limited or biased training data and the lack of suitable inductive bias in models. Such biases also affect fact verification Symmetric (Counterfactual) Evide... | Fine-tuning broadly refers to approaches where a model is initially trained on one dataset and then further improved by training on another. We refer to these datasets as fine-tuning training and test data as FT-train and FT-test respectively. This technique is commonly used to mitigate model biases Elastic Weight Cons... | 782 | 558 | 782 |
How Helpful is Inverse Reinforcement Learning for Table-to-Text Generation? | Existing approaches for the Table-to-Text task suffer from issues such as missing information, hallucination and repetition. Many approaches to this problem use Reinforcement Learning (RL), which maximizes a single manually defined reward, such as BLEU. In this work, we instead pose the Table-to-Text task as Inverse Re... | Table -to-Text generation focuses on explaining tabular data in natural language. This is increasingly relevant due to the vast amounts of tabular data created in domains including e-commerce, healthcare and industry (for example, infoboxes in Wikipedia, tabular product descriptions in online shopping sites, etc.). Tab... | The training data for this task consists of pairs of tables and corresponding natural language descriptions, as shown in Figure We pose Table-to-Text under the IRL framework where we aim to jointly learn a policy for generating description from the table and the underlying composite reward function. At the core of our ... | 807 | 957 | 807 |
Making Use of Latent Space in Language GANs for Generating Diverse Text without Pre-training | Generating diverse texts is an important factor for unsupervised text generation. One approach is to produce the diversity of texts conditioned by the sampled latent code. Although several generative adversarial networks (GANs) have been proposed thus far, these models still suffer from mode-collapsing if the models ar... | Generative adversarial networks (GANs) For language GANs, the diversity of the generated texts is an important evaluation metric. There are mainly two approaches to produce the diversity of texts by the generative models. One approach, which includes SeqGAN In this paper, we propose a GAN model that aims to improve the... | The language GANs, in which the generator and discriminator optimize their objective functions in an adversarial manner to generate realistic texts, have two main perspectives. As the first perspective, a reinforcement learning approach is used for optimizing the generator. SeqGAN As the second perspective, the model i... | 1,104 | 2,146 | 1,104 |
Discovering Dialogue Slots with Weak Supervision | Task-oriented dialogue systems typically require manual annotation of dialogue slots in training data, which is costly to obtain. We propose a method that eliminates this requirement: We use weak supervision from existing linguistic annotation models to identify potential slot candidates, then automatically identify do... | Task-oriented dialogue systems typically use annotation based on slots to represent the meaning of user utterances Getting raw data for dialogue system training is not difficult, especially if we restrict the target domain. A requirement for dialogue state labels makes this process much more costly. However, both tradi... | Figure Although the annotation is descriptive, it contains concepts irrelevant for the domain under consideration. Our method selects only relevant slot candidates (depicted in blue). Slots discovered by our approach can then be used to design or adapt the database backend for the target domain. Our contributions can b... | 628 | 1,115 | 628 |
RLET: A Reinforcement Learning Based Approach for Explainable QA with Entailment Trees | Interpreting the reasoning process from questions to answers poses a challenge in approaching explainable QA. A recently proposed structured reasoning format, entailment tree, manages to offer explicit logical deductions with entailment steps in a tree structure. To generate entailment trees, prior single pass sequence... | Reasoning over explicitly given knowledge and generating detailed deduction steps are important challenges towards the goal of automated reasoning in AI community One line of previous work considers the entailment trees as linearised sequences and adopt sequence-to-sequence (Seq2Seq) models to generate the entire reaso... | Our goal is to provide a step-by-step reasoning process for commonsense science questions, with prior knowledge of the question, the correct answer and a set of fact sentences. We first describe the task formulation and then explain each part in detail. We formulate the reasoning process as an entailment tree construct... | 1,137 | 1,877 | 1,137 |
Clique-Based Clustering for improving Named Entity Recognition systems | We propose a system which builds, in a semi-supervised manner, a resource that aims at helping a NER system to annotate corpus-specific named entities. This system is based on a distributional approach which uses syntactic dependencies for measuring similarities between named entities. The specificity of the presented ... | In Information Extraction domain, named entities (NEs) are one of the most important textual units as they express an important part of the meaning of a document. Named entity recognition (NER) is not a new domain (see MUC • intra-annotation ambiguity: Wikipedia lists more than 25 cities named Oxford in the world • sys... | Given a corpus, the main objectives of our system are: to detect potential NEs; to compute the possible annotations for each NE and then; to annotate each occurrence of these NEs with the right annotation by analyzing its local context. We assume that this corpus dependent approach allows an easier NE annotation. Indee... | 578 | 1,436 | 578 |
TwittIrish: A Universal Dependencies Treebank of Tweets in Modern Irish | Modern Irish is a minority language lacking sufficient computational resources for the task of accurate automatic syntactic parsing of usergenerated content such as tweets. Although language technology for the Irish language has been developing in recent years, these tools tend to perform poorly on user-generated conte... | Irish is a minority language spoken mostly in small communities in Ireland called 'Gaeltachtaí' User-generated content (UGC), such as tweets, is a valuable, highly available resource for training syntactic parsers that can accurately process social media text. UGC is a genre with features different from those of both s... | We use the following resources: Indigenous Tweets (IT) Lynn Twitter Corpus (LTC) Irish Universal Dependencies Treebank (IUDT) gaBERT We combined 700 POS-tagged tweets from the LTC with 166 tweets more recently crawled by IT in order to leverage previous linguistic annotations while also including newer tweets. This inv... | 962 | 1,528 | 962 |
Linking Surface Facts to Large-Scale Knowledge Graphs | Open Information Extraction (OIE) methods extract facts from natural language text in the form of ("subject"; "relation"; "object") triples. These facts are, however, merely surface forms, the ambiguity of which impedes their downstream usage; e.g., the surface phrase "Michael Jordan" may refer to either the former bas... | Open Information Extraction (OIE) methods extract surface ("subject"; "relation"; "object")triples from natural language text in a schema-free manner Knowledge Graphs (KGs), on the other hand, are inventories of canonical facts in the form of (subject; predicate; object)-triples, where each slot is a unique (i.e., unam... | For a given surface-form OIE triple t 1 = ("s"; "r"; "o"), the goal is to link each slot to a canonical concept in a KG (if the corresponding concept exists in the KG): "s" → e 1 ∈ E; "r" → p ∈ P; "o" → e 2 ∈ E, with E and P as the (fixed) sets of KG entities and predicates. Importantly, our problem definition (and con... | 1,285 | 1,703 | 1,285 |
BOTEVAL: Facilitating Interactive Human Evaluation | Following the rapid progress in natural language processing (NLP) models, language models are applied to increasingly more complex interactive tasks such as negotiations and conversation moderations. Having human evaluators directly interact with these NLP models is essential for adequately evaluating the performance o... | As natural language processing (NLP) models become more versatile with the recent advances of language models and their instruction-tuned counterparts As noted by To facilitate accurate human evaluations of complex interactive tasks, we developed BOTEVAL, 1 a comprehensive evaluation toolkit that focuses on enabling hu... | BOTEVAL is a web application that provides an evaluation interface, what the human evaluators (i.e., crowdsource workers) see (Section 2.1), and an administrator dashboard, what the administrator uses to manage the evaluation task and evaluators (Section 2.2). We recommend that the bots that evaluators interact with ar... | 1,023 | 1,310 | 1,023 |
Contrastive Analysis with Predictive Power: Typology Driven Estimation of Grammatical Error Distributions in ESL | This work examines the impact of crosslinguistic transfer on grammatical errors in English as Second Language (ESL) texts. Using a computational framework that formalizes the theory of Contrastive Analysis (CA), we demonstrate that language specific error distributions in ESL writing can be predicted from the typologic... | The study of cross-linguistic transfer, whereby properties of a native language influence performance in a foreign language, has a long tradition in Linguistics and Second Language Acquisition (SLA). Much of the linguistic work on this topic was carried out within the framework of Contrastive Analysis (CA), a theoretic... | Cross linguistic-transfer was extensively studied in SLA, Linguistics and Psychology Computational work touching on crosslinguistic transfer was mainly conducted in relation to the Native Language Identification (NLI) task, in which the goal is to determine the native language of the author of an ESL text. Much of this... | 956 | 3,288 | 956 |
Exploiting domain-slot related keywords description for Few-Shot Cross-Domain Dialogue State Tracking | Collecting dialogue data with domain-slotvalue labels for dialogue state tracking (DST) could be a costly process. In this paper, we propose a novel framework based on domain-slot related description to tackle the challenge of few-shot cross-domain DST. Specifically, we design an extraction module to extract domainslot... | Dialogue state tracking (DST) is an essential component in a task-oriented dialogue system. It aims to keep track of users' domains, intents and slots information at each turn of the conversation, which helps to provide sufficient information for selecting the next system operation (1) Modular methods | In this work, we proposed a simple but efficient framework named Domain-slot Related Information Awareness method (DRIA) based on the domain-slot related keywords extraction module and a random sampling strategy. Specifically, for the extraction module, we first use TF-IDF algorithm Our contributions are summarized as ... | 920 | 302 | 920 |
Using support vector machines and state-of-the-art algorithms for phonetic alignment to identify cognates in multi-lingual wordlists | Most current approaches in phylogenetic linguistics require as input multilingual word lists partitioned into sets of etymologically related words (cognates). Cognate identification is so far done manually by experts, which is time consuming and as of yet only available for a small number of well-studied language famil... | Computational historical linguistics is a relatively young sub-discipline of computational linguistics which uses computational methods to uncover how the world's 7 000 human languages have developed into their current shape. The discipline has made great strides in recent years. Exciting progress has been made with re... | Cognate detection is a partitioning task: a clustering task which does not necessarily assume a hierarchy. An early approach An alternative family of approaches to cognate detection circumvents this problem by first calculating distances or similarities between pairs of words in the data, and then feeding those scores ... | 1,108 | 1,567 | 1,108 |
Representation and Generation of Machine Learning Test Functions | Writing tests for machine learning (ML) code is a crucial step towards ensuring the correctness and reliability of ML software. At the same time, Large Language Models (LLMs) have been adopted at a rapid pace for various code generation tasks, making it a natural choice for many developers who need to write ML tests. H... | As AI and ML become more and more integrated into everyday processes, ensuring the quality and reliability of these ML models is mandatory, and a critical part of ensuring ML models' performance in production is having good, representative test cases. Traditionally, these tests have been written by engineers and progra... | The focus on learning distributed representations of code forms the groundwork of our research. We draw from It's also important to mention the effort on benchmarking datasets like CodeSearchNet Language Models on Source Code Substantial research has been invested in revealing the power of LLMs in dealing with code-rel... | 947 | 1,810 | 947 |
Early Detection of Sexual Predators in Chats | An important risk that children face today is online grooming, where a so-called sexual predator establishes an emotional connection with a minor online with the objective of sexual abuse. Prior work has sought to automatically identify grooming chats, but only after an incidence has already happened in the context of ... | Online grooming denotes the process where a socalled sexual predator establishes an emotional connection with a minor online to systematically solicit and exploit them for sexual purposes The problem of detecting whether or not a child is being groomed by a predator is called sexual predator detection We believe that i... | We introduce the task of early sexual predator • We introduce the problem of eSPD and formally define it. • We survey available datasets, analyze their limitations, and build a new combined dataset called PANC as a best-effort for evaluating eSPD. • We propose a task setup to evaluate eSPD, focusing on the trade-off be... | 1,004 | 1,004 | 1,004 |
Semantic Frame Induction with Deep Metric Learning | Recent studies have demonstrated the usefulness of contextualized word embeddings in unsupervised semantic frame induction. However, they have also revealed that generic contextualized embeddings are not always consistent with human intuitions about semantic frames, which causes unsatisfactory performance for frame ind... | Semantic frames are knowledge resources that reflect human intuitions about various concepts such as situations and events. One of the most representative semantic frame resources is FrameNet | Example sentence FILLING (1) She covered her mouth with her hand. (2) I filled a notebook with my name. (3) You can embed graphs in your worksheet. (4) He parked the car at the hotel. (5) Volunteers removed grass from the marsh. (6) They'd drained the drop from the teapot. TOPIC (7) Each database will cover a specific ... | 1,091 | 191 | 1,091 |
MQAG: Multiple-choice Question Answering and Generation for Assessing Information Consistency in Summarization | State-of-the-art summarization systems can generate highly fluent summaries. These summaries, however, may contain factual inconsistencies and/or information not present in the source. Hence, an important component of assessing the quality of summaries is to determine whether there is information consistency between th... | The objective of summary evaluation is to quantify the quality of summaries, either on a relative or an absolute scale. Accurate and reliable automatic summary evaluation systems are useful to researchers, as they provide an easy and cheap way to compare new summarization models to existing ones. Although current summa... | Standard summary evaluation metrics such as ROUGE Textual overlap scores n-gram based metrics, including BLEU where T x and T y are relation triples extracted from the source and the summary, respectively. Simulated data, such as real or fake summaries created by pre-defined transformations, have been used to train cl... | 1,272 | 1,398 | 1,272 |
Diversity Enhanced Narrative Question Generation for StoryBooks | Question generation (QG) from a given context can enhance comprehension, engagement, assessment, and overall efficacy in learning or conversational environments. Despite recent advancements in QG, the challenge of enhancing or measuring the diversity of generated questions often remains unaddressed. In this paper, we i... | Question generation (QG), focusing on the questions derived from specific text passages or documents, plays an integral role in a wide array of domains. It improves question answering (QA) systems The importance of generating and evaluating multiple questions becomes evident when we examine the creation process of QA d... | Based on given contents, question generation aims to generate natural language questions, where the generated questions are able to be addressed with the given contents. After neural approaches took over a large proportion in QG In natural language generation (NLG), generating outputs that are not only correct but also... | 945 | 2,326 | 945 |
Unsupervised Multilingual Sentence Embeddings for Parallel Corpus Mining | Existing models of multilingual sentence embeddings require large parallel data resources which are not available for low-resource languages. We propose a novel unsupervised method to derive multilingual sentence embeddings relying only on monolingual data. We first produce a synthetic parallel corpus using unsupervise... | Parallel corpora constitute an essential training data resource for machine translation as well as other cross-lingual NLP tasks. However, large parallel corpora are only available for a handful of language pairs while the rest relies on semi-supervised or unsupervised methods for training. Since monolingual data are g... | Related research comprises supervised methods to model multilingual sentence embeddings and unsupervised methods to model multilingual word embeddings which can be aggregated into sentences. Furthermore, our approach is closely related to the recent research in cross-lingual language model (LM) pretraining. Supervised ... | 734 | 1,072 | 734 |
Bias in Opinion Summarisation from Pre-training to Adaptation: A Case Study in Political Bias | Opinion summarisation aims to summarise the salient information and opinions presented in documents such as product reviews, discussion forums, and social media texts into short summaries that enable users to effectively understand the opinions therein. Generating biased summaries has the risk of potentially swaying pu... | Opinion summarisation aims to condense the opinions presented in the source documents into a summary so that readers can effectively comprehend the opinions in the source documents using input data such as product reviews A summarisation model's output will reflect any biases inherited from the training data. Pre-train... | Opinion summarisation is a task to summarise user opinions expressed in different online media, such as product reviews, social media conversations, and online discussion forums. There are two primary types of models: extractive -selecting salient sentences from input documents Existing studies of bias in opinion summa... | 1,091 | 1,840 | 1,091 |
A Holistic Approach to Reference-Free Evaluation of Machine Translation | Traditional machine translation evaluation relies on references written by humans. While reference-free evaluation gets rid of the constraints of labor-intensive annotations, it can pivot easily to new domains and is more scalable. In this paper, we propose a referencefree evaluation approach that characterizes evaluat... | Machine translation evaluation has conventionally relied on reference, where outputs are compared against translations written by humans. This is in contrast to the reference-free manner in which translation quality is directly assessed with the source text. Reference-free evaluation The history of reference-free evalu... | Reference-free evaluation of MT can be characterized as two aspects: (1) fluency: how well it conforms to normal human language usage; and (2) faithfulness: how well the translated text reflects the source data. We assess faithfulness at different granularity: word level and sentence level. Figure We explore a data aug... | 884 | 1,357 | 884 |
FLEEK: Factual Error Detection and Correction with Evidence Retrieved from External Knowledge | Detecting factual errors in textual information, whether generated by large language models (LLM) or curated by humans, is crucial for making informed decisions. LLMs' inability to attribute their claims to external knowledge and their tendency to hallucinate makes it difficult to rely on their responses. Humans, too, ... | While textual information offers a convenient and efficient means of communication, it is critical to acknowledge its potential for misuse or unintended consequences. False or misleading information spreads easily over online platforms Previous works In this work, we present FLEEK (FactuaL Error detection and correctio... | Figure Taylor Swift is 30 years old In this work, we define a fact as a unit of information that (1) describes a certain entity or (2) captures the relation between two entities (3) describes an event. Each fact consists of a subject, a predicate, and at least one object. We use the semi-structured triple format to rep... | 894 | 1,231 | 894 |
PaperMage: A Unified Toolkit for Processing, Representing, and Manipulating Visually-Rich Scientific Documents | Despite growing interest in applying natural language processing (NLP) and computer vision (CV) models to the scholarly domain, scientific documents remain challenging to work with. They're often in difficult-to-use PDF formats, and the ecosystem of models to process them is fragmented and incomplete. We introduce pape... | Research papers and textbooks are central to the scientific enterprise, and there is increasing interest in developing new tools for extracting knowledge from these visually-rich documents. Recent research has explored, for example, AI-powered reading support for math symbol definitions However, this type of NLP resear... | Processing visually-rich documents like scientific documents requires a joint understanding of both visual and textual information. In practice, this often requires combining different models into complex processing pipelines. For example, GRO-BID While aforementioned software tools use CRF or BiLSTM-based models, Tran... | 952 | 1,647 | 952 |
Improving Adversarial Text Generation by Modeling the Distant Future | Auto-regressive text generation models usually focus on local fluency, and may cause inconsistent semantic meaning in long text generation. Further, automatically generating words with similar semantics is challenging, and hand-crafted linguistic rules are difficult to apply. We consider a text planning scheme and pres... | Text generation is an important area of investigation within machine learning. Recent work has shown excellent performance on a number of tasks, by combining reinforcement learning (RL) and generative models. Example applications include image captioning For RL-based text generation, most existing works rely on a model... | Text Generation Model Text generation models learn to generate a sentence Y = (y 1 , . . . , y T ) of length T , possibly conditioned on some context X. Here each y t is a token from vocabulary A. Starting from the initial state s 0 , a recurrent neural network (RNN) produces a sequence of states (s 1 , . . . , s T ) g... | 700 | 1,257 | 700 |
Handling Japanese Homophone Errors in Revision Support System for Japanese Texts; REVISE | Japanese texts frequently suffer from the homophone errors caused by the KANA-KANJI conversion needed to input the text. It is critical, therefore, for Japanese revision support systems to detect and to correct homophone errors. This paper proposes a method for detecting and correcting Japanese homophone errors in comp... | We have been using morphological analysis to develop REVISE, a revision support system that corrects Japanese input errors Most Japanese texts are made with Japanese word processors. As Japanese texts consist of phonograms, KANA, and ideograms, KANJI, lapanese word processors always use KANA-KANJI conversion in which K... | Key terms used in this paper are defined as follows: • Japanese compound noun; A noun that consists of several nouns, none of which have JOSHI (i.e. Japanese postpositions). A word that sounds the same as another but has different spelling (i.e. KANJI sequence) and meaning. • Homophone error; An error that occurs when ... | 988 | 1,320 | 988 |
INTERFAIR: Debiasing with Natural Language Feedback for Fair Interpretable Predictions | Debiasing methods in NLP models traditionally focus on isolating information related to a sensitive attribute (e.g. gender or race). We instead argue that a favorable debiasing method should use sensitive information 'fairly,' with explanations, rather than blindly eliminating it. This fair balance is often subjective ... | Debiasing human written text is an important scientific and social problem that has been investigated by several recent works However, a user can potentially further tune the model's belief on the bias, leading to a correct prediction while minimally using biased information. While interactive NLP models recently focus... | An interpretable debiasing algorithm produces a rationale along with a prediction of the original task to expose the amount of bias or sensitive information used. Precisely, a rationale is the minimal and sufficient part of the input responsible for the prediction. For text input, let the predictive input tokens for th... | 926 | 793 | 926 |
Coarse-to-Fine Decoding for Neural Semantic Parsing | Semantic parsing aims at mapping natural language utterances into structured meaning representations. In this work, we propose a structure-aware neural architecture which decomposes the semantic parsing process into two stages. Given an input utterance, we first generate a rough sketch of its meaning, where low-level i... | Semantic parsing maps natural language utterances onto machine interpretable meaning representations (e.g., executable queries or logical forms). The successful application of recurrent neural networks to a variety of NLP tasks In this work, we propose to decompose the decoding process into two stages. The first decode... | Length Example GEO 7.6 13.7 6.9 x : which state has the most rivers running through it? y : (argmax $0 (state:t $0) (count $1 (and (river:t $1) (loc:t $1 $0)))) a : (argmax#1 state:t@1 (count#1 (and river:t@1 loc:t@2 ) ) ) ATIS 11.1 21.1 9.2 x : all flights from dallas before 10am y : (lambda $0 e (and (flight $0) (fro... | 729 | 1,269 | 729 |
Stage-wise Fine-tuning for Graph-to-Text Generation | Graph-to-text generation has benefited from pre-trained language models (PLMs) in achieving better performance than structured graph encoders. However, they fail to fully utilize the structure information of the input graph. In this paper, we aim to further improve the performance of the pre-trained language model by p... | In the graph-to-text generation task Table We explore the proposed stage-wise fine-tuning and structure-preserving embedding strategies for graph-to-text generation task on WebNLG corpus | Given an RDF graph with multiple relations our goal is to generate a text faithfully describing the input graph. We represent each relation with a triple (s i , r i , o i ) ∈ G for i ∈ {1, ..., n}, where s i , r i , and o i are natural language phrases that represent the subject, type, and object of the relation, respe... | 836 | 186 | 836 |
RASAT: Integrating Relational Structures into Pretrained Seq2Seq Model for Text-to-SQL | Relational structures such as schema linking and schema encoding have been validated as a key component to qualitatively translating natural language into SQL queries. However, introducing these structural relations comes with prices: they often result in a specialized model structure, which largely prohibits using lar... | Text-to-SQL is the task that aims at translating natural language questions into SQL queries. Since it could significantly break down barriers for nonexpert users to interact with databases, it is among the most important semantic parsing tasks that are of practical importance Various types of relations have been intro... | Early works usually exploit a sketch-based slotfilling method that uses different modules to predict the corresponding part of SQL. These methods decompose the SQL generation task into several independent sketches and use different classifiers to predict corresponding part, such as SQLNet Faced with the multi-table and... | 1,046 | 2,289 | 1,046 |
Zero-shot Neural Passage Retrieval via Domain-targeted Synthetic Question Generation | A major obstacle to the wide-spread adoption of neural retrieval models is that they require large supervised training sets to surpass traditional term-based techniques, which are constructed from raw corpora. In this paper, we propose an approach to zero-shot learning for passage retrieval that uses synthetic question... | Recent advances in neural retrieval have led to advancements on several document, passage and knowledge-base benchmarks | Retrieval Model Rescoring Model Figure Another consideration is that BM25 is often high quality The focus of the present work is methods for building neural models for first-stage passage retrieval for large collections of documents. While rescoring models are key components to any retrieval system, they are out of the... | 950 | 119 | 950 |
Coherence-based Modeling of Clinical Concepts Inferred from Heterogeneous Clinical Notes for ICU Patient Risk Stratification | In hospitals, critical care patients are often susceptible to various complications that adversely affect their morbidity and mortality. Digitized patient data from Electronic Health Records (EHRs) can be utilized to facilitate risk stratification accurately and provide prioritized care. Existing clinical decision supp... | Until recently, the healthcare industry had an inclination towards conservative approaches for the treatment and diagnosis of patients, resulting in less patient-centric and imprecise assessments Pat is 83 yo F w/PMHx for CLL and hypotens, who was admited for an elective total hip arthroplasty for persistent hip pain. ... | In this section, we discuss in detail, the Natural Language Processing (NLP) pipeline designed to facilitate multi-label ICD-9 code group prediction, and the same is depicted in Figure MIMIC-III (v1.4) is a publicly available large healthcare database with comprehensive medical data of over 40, 000 ICU patients. The he... | 1,721 | 3,132 | 1,721 |
Holistic Inter-Annotator Agreement and Corpus Coherence Estimation in a Large-scale Multilingual Annotation Campaign | In this paper we report on the complexity of persuasion technique annotation in the context of a large multilingual annotation campaign involving 6 languages and approximately 40 annotators. We highlight the techniques that appear to be difficult for humans to annotate and elaborate on our findings on the causes of thi... | In the recent years we have observed an emergence of automated tools for facilitating online media analysis for better understanding of the presented narratives around certain topics across countries, and to identify manipulative, deceptive and propagandistic content. Developing such tools requires annotated data of hi... | Persuasion detection in text is related to work on propaganda detection. The work in the latter area initially focused on document-level analysis and predictions, e.g., In parallel, other efforts focused on the detection of specific persuasion techniques. Various related shared tasks on the detection of persuasion tech... | 936 | 4,066 | 936 |
SPANNER: Named Entity Re-/Recognition as Span Prediction | Recent years have seen the paradigm shift of Named Entity Recognition (NER) systems from sequence labeling to span prediction. Despite its preliminary effectiveness, the span prediction model's architectural bias has not been fully understood. In this paper, we first investigate the strengths and weaknesses when the sp... | The rapid evolution of neural architectures However, despite the success of span predictionbased systems, as a relatively newly-explored framework, the understanding of its architectural bias has not been fully understood so far. For example, what are the complementary advantages compared with SEQLAB frameworks and how... | NER is frequently formulated as a sequence labeling (SEQLAB) problem To make a comprehensive evaluation, in this paper, we use multiple NER datasets that cover different domains and languages. CoNLL-2003 2 (Sang and De Meulder, 2003) covers two different languages: English and German. Here, we only consider the English... | 946 | 2,043 | 946 |
Boosting Neural Machine Translation with Similar Translations | This paper explores data augmentation methods for training Neural Machine Translation to make use of similar translations, in a comparable way a human translator employs fuzzy matches. In particular, we show how we can simply feed the neural model with information on both source and target sides of the fuzzy matches, w... | For decades, the localization industry has been proposing Fuzzy Matching technology in CAT tools allowing the human translator to visualize one or several fuzzy matches from translation memory when translating a sentence leading to higher productivity and consistency With improving machine translation technology 1 and ... | A translation memory (TM) is a database that stores translated segments composed of a source and its corresponding translations. It is mostly used to match up previous translations to new content that is similar to content translated in the past. Assuming that we translated the following English sentence into French: [... | 1,066 | 1,533 | 1,066 |
EmoNoBa: A Dataset for Analyzing Fine-Grained Emotions on Noisy Bangla Texts | For low-resourced Bangla language, works on detecting emotions on textual data suffer from size and cross-domain adaptability. In our paper, we propose a manually annotated dataset of 22,698 Bangla public comments from social media sites covering 12 different domains such as Personal, Politics, and Health, labeled for ... | Identifying emotions has helped find solutions to numerous problems for English text, namely retrieving emotion from suicide notes Bangla is the sixth most spoken language globally 2 and is the native language of Bangladesh. † First and second authors contributed equally 1 Data and code available at | [B] এইরকম েশা-অফ হাজার বার েদখেত চাই। Few datasets have been made public for detecting emotion in a low-resourced Bangla language In this paper, we aim to create a multi-label emotion dataset of noisy textual data collected from social media on various topics. We use the Junto emotion wheel • We propose EmoNoBa dataset... | 717 | 300 | 717 |
What's in a Name? Entity Type Variation across Two Biomedical Subdomains | There are lexical, syntactic, semantic and discourse variations amongst the languages used in various biomedical subdomains. It is important to recognise such differences and understand that biomedical tools that work well on some subdomains may not work as well on others. We report here on the semantic variations that... | Biomedical information extraction efforts in the past decade have focussed on fundamental tasks needed to create intelligent systems capable of improving search engine results and easing the work of biologists. More specifically, researchers have concentrated mainly on named entity recognition, mapping them to concepts... | Harris (1968) introduced a formalisation of the notion of sublanguage, which was defined as a subset of general language. According to this theory, it is possible to process specialised languages, since they have a structure that can be expressed in a computable form. More recently, several works on the study of biomed... | 748 | 1,368 | 748 |
Optimizing Retrieval-augmented Reader Models via Token Elimination | Fusion-in-Decoder (FiD) is an effective retrieval-augmented language model applied across a variety of open-domain tasks, such as question answering, fact checking, etc. In FiD, supporting passages are first retrieved and then processed using a generative model (Reader), which can cause a significant bottleneck in deco... | The task of Open-Domain Question Answering (ODQA) The retriever-reader architecture has been widely-used and adopted for ODQA tasks of the reading process. In order to assess ODQA methods, There has been rapid and remarkable progress in retriever-reader systems for solving ODQA tasks using a generative approach Previou... | In a retriever-reader system, the reader, which is typically a language model, receives a query along with a collection of passages, where each passage often consists of a title and a context. Additionally, we are provided with the ground truth, which can be an expected answer or a gold passage that is most relevant to... | 813 | 1,894 | 813 |
Zero-Shot Entity Linking by Reading Entity Descriptions | We present the zero-shot entity linking task, where mentions must be linked to unseen entities without in-domain labeled data. The goal is to enable robust transfer to highly specialized domains, and so no metadata or alias tables are assumed. In this setting, entities are only identified by text descriptions, and mode... | Entity linking systems have achieved high performance in settings where a large set of disambiguated mentions of entities in a target entity dictionary is available for training. Such systems typically use powerful resources such as a high-coverage alias table, structured data, and linking frequency statistics. For exa... | We first review standard entity linking task definitions and discuss assumptions made by prior systems. We then define the zero-shot entity linking task and discuss its relationship to prior work. Entity linking (EL) is the task of grounding entity mentions by linking them to entries in a given database or dictionary o... | 976 | 2,287 | 976 |
End-to-End Single-Channel Speaker-Turn Aware Conversational Speech Translation | Conventional speech-to-text translation (ST) systems are trained on single-speaker utterances, and they may not generalize to real-life scenarios where the audio contains conversations by multiple speakers. In this paper, we tackle single-channel multi-speaker conversational ST with an end-to-end and multi-task trainin... | Speech translation (ST) has seen wide adoption in commercial products and the research community 2017, inter alia) have recently gained increasing interest and popularity thanks to their simple architecture, less error propagation Despite significant recent advances in E2E-ST In this work, we tackle the more challengin... | Joint ST & ASR Modeling Recent works in ST have leveraged ASR training data to improve translation quality. In principle, joint ASR and ST modeling Conversational Speech Translation Work on conversational ST In this work, we report results on the Fisher-CALLHOME corpus Speaker-Turn and Cross-Talk in ASR Speakerturns an... | 1,146 | 728 | 1,146 |
Optimal Transport for Unsupervised Hallucination Detection in Neural Machine Translation | Neural machine translation (NMT) has become the de-facto standard in real-world machine translation applications. However, NMT models can unpredictably produce severely pathological translations, known as hallucinations, that seriously undermine user trust. It becomes thus crucial to implement effective preventive stra... | Neural machine translation (NMT) has achieved tremendous success In this work, we focus on leveraging the crossattention mechanism to develop a novel hallucination detector. This mechanism is responsible for selecting and combining the information contained in the source sequence that is relevant to retain during trans... | A NMT model M defines a probability distribution p θ (y|x) over an output space of hypotheses Y conditioned on a source sequence x contained in an input space X . In this work, we focus on models parameterized by an encoder-decoder transformer model The first-order Wasserstein distance between two arbitrary probability... | 1,085 | 1,584 | 1,085 |
When a Good Translation is Wrong in Context: Context-Aware Machine Translation Improves on Deixis, Ellipsis, and Lexical Cohesion | Though machine translation errors caused by the lack of context beyond one sentence have long been acknowledged, the development of context-aware NMT systems is hampered by several problems. Firstly, standard metrics are not sensitive to improvements in consistency in document-level translations. Secondly, previous wor... | With the recent rapid progress of neural machine translation (NMT), translation mistakes and inconsistencies due to the lack of extra-sentential context are becoming more and more noticeable among otherwise adequate translations produced by standard context-agnostic NMT systems A context-agnostic NMT system would often... | We begin with a human study, in which we: 1. identify cases when good sentence-level translations are not good when placed in context of each other, 2. categorize these examples according to the phenomena leading to a discrepancy in translations of consecutive sentences. The test sets introduced in Section 3 will then ... | 1,143 | 1,359 | 1,143 |
Maieutic Prompting: Logically Consistent Reasoning with Recursive Explanations | Pre-trained language models (LMs) struggle with consistent reasoning; recently, prompting LMs to generate explanations that self-guide the inference has emerged as a promising direction to amend this. However, these approaches are fundamentally bounded by the correctness of explanations, which themselves are often nois... | Following the remarkable success of few-shot prompting over large language models (e.g. | Input Prompt Q: Captain Kirk is part of Star Wars? A: Captain Kirk is a character in Star Trek. Therefore, the answer is False. Output A: There are female mayors. Therefore, the answer is True. Smoke is not the source of fire? Smoke is a result of fire. Therefore, the statement is False. One is a number that comes befo... | 1,114 | 87 | 1,114 |
On the Practical Computational Power of Finite Precision RNNs for Language Recognition | While Recurrent Neural Networks (RNNs) are famously known to be Turing complete, this relies on infinite precision in the states and unbounded computation time. We consider the case of RNNs with finite precision whose computation time is linear in the input length. Under these limitations, we show that different RNN va... | Recurrent Neural Network (RNNs) emerge as very strong learners of sequential data. A famous result by In this work we restrict ourselves to inputbound recurrent neural networks with finiteprecision states (IBFP-RNN), trained using backpropagation. This class of networks is likely to coincide with the networks one can e... | An RNN is a parameterized function R that takes as input an input vector x t and a state vector h t-1 and returns a state vector h t : The RNN is applied to a sequence x 1 , ..., x n by starting with an initial vector h 0 (often the 0 vector) and applying R repeatedly according to equation (1). Let Σ be an input vocabu... | 694 | 1,522 | 694 |
Evaluating Explanation Methods for Neural Machine Translation | Recently many efforts have been devoted to interpreting the black-box NMT models, but little progress has been made on metrics to evaluate explanation methods. Word Alignment Error Rate can be used as such a metric that matches human understanding, however, it can not measure explanation methods on those target words t... | Neural machine translation (NMT) has witnessed great success during recent years Generally speaking, we recognize two orthogonal dimensions for evaluating the explanation methods: i) how much the pattern (such as source words) extracted by an explanation method matches human understanding on predicting a target word; o... | Suppose Most NMT literature models the following conditional probability P (y | x) in an encoder-decoder fashion: where y <t = {y 1 , • • • , y t-1 } denotes a prefix of y with length t -1, and s t is the decoding state vector of timestep t. In the encoding stage, the encoder of a NMT model transforms the source senten... | 904 | 2,211 | 904 |
SARAL: A Low-Resource Cross-Lingual Domain-Focused Information Retrieval System for Effective Rapid Document Triage | With the increasing democratization of electronic media, vast information resources are available in less-frequently-taught languages such as Swahili or Somali. That information, which may be crucially important and not available elsewhere, can be difficult for monolingual English speakers to effectively access. In thi... | The task of searching for a needle of relevant information in a haystack of documents is not as daunting as in previous eras, thanks to decades of information retrieval research progress. Most of us engage in this behavior daily when we search the web. Powerful IR algorithms choose the most likely matches for our queri... | ) is a Hindi word which can be translated as ingenious or simple, depending on the relevant context. 1. SEARCHER, a novel CLIR approach designed for low-resource conditions that relies on the construction of a shared semantic space learned from bitext and monolingual corpora 2. An intuitive snippet extraction and prese... | 854 | 2,558 | 854 |
Composing extended top-down tree transducers * | A composition procedure for linear and nondeleting extended top-down tree transducers is presented. It is demonstrated that the new procedure is more widely applicable than the existing methods. In general, the result of the composition is an extended top-down tree transducer that is no longer linear or nondeleting, bu... | Tree-based translation models such as synchronous tree substitution grammars In this way we model the word drop by an ln-XTOP M and reordering by an ln-XTOP N . The syntactic properties of linearity and nondeletion yield nice algorithmic properties, and the mod- ular approach is desirable for better design and parametr... | Our trees have labels taken from an alphabet Σ of symbols, and in addition, leaves might be labeled by elements of the countably infinite set X = {x 1 , x 2 , . . . } of formal variables. Formally, for every V ⊆ X the set T Σ (V ) of Σ-trees with V -leaves is the smallest set such that V ⊆ T Σ (V ) and σ(t 1 , . . . , ... | 410 | 1,332 | 410 |
Action-Sensitive Phonological Dependencies | This paper defines a subregular class of functions called the tier-based synchronized strictly local (TSSL) functions. These functions are similar to the the tier-based inputoutput strictly local (TIOSL) functions, except that the locality condition is enforced not on the input and output streams, but on the computatio... | The subregular program in phonology seeks to define subclasses of the regular languages and finitestate functions that describe attested phonotactic constraints and phonological processes. These subclasses provide a natural framework for typological classification of linguistic phenomena while allowing for the developm... | As usual, N denotes the set of nonnegative integers. Σ and Γ denote finite alphabets not including the left and right word boundary symbols and , respectively. The length of a string x is denoted by |x|, and λ denotes the empty string. Alphabet symbols are identified with strings of length 1, and individual strings are... | 624 | 1,931 | 624 |
Using Human Attention to Extract Keyphrase from Microblog Post | This paper studies automatic keyphrase extraction on social media. Previous works have achieved promising results on it, but they neglect human reading behavior during keyphrase annotating. The human attention is a crucial element of human reading behavior. It reveals the relevance of words to the main topics of the ta... | Rapidly growth of user-generated content on social media has far outpaced human beings' reading and understanding capacity. Keyphrase extraction is one of the technologies that can organize this massive content. A keyphrase consists of one or more salient words, which represents the main topics of a document. It has a ... | Recently, keyphrase extraction technologies have been extended to social media The open source eye-tracking corpus of natural reading include the Dundee corpus Formally, given a target microblog post x i formulated as word sequence where y i,w indicates whether x i,w is part of a keyphrase. As shown in Figure where h i... | 877 | 1,083 | 877 |
Learning to Exploit Structured Resources for Lexical Inference | Massive knowledge resources, such as Wikidata, can provide valuable information for lexical inference, especially for proper-names. Prior resource-based approaches typically select the subset of each resource's relations which are relevant for a particular given task. The selection process is done manually, limiting th... | Recognizing lexical inference is an important component in semantic tasks. Various lexicalsemantic relations, such as synonomy, classmembership, part-of, and causality may be used to infer the meaning of one word from another, in order to address lexical variability. For instance, a question answering system asked "whi... | WordNet One approach looks for chains of these predefined relations While there is a broad consensus that synonyms entail each other (elevator ↔ lif t) and hyponyms entail their hypernyms (cat → animal), other relations, such as meronymy, are not agreed Overall, there is no principled way to select the subset of releva... | 724 | 2,127 | 724 |
Selective Labeling: How to Radically Lower Data-Labeling Costs for Document Extraction Models | Building automatic extraction models for visually rich documents like invoices, receipts, bills, tax forms, etc. has received significant attention lately. A key bottleneck in developing extraction models for new document types is the cost of acquiring the several thousand highquality labeled documents that are needed ... | Visually rich documents such as invoices, receipts, paystubs, insurance statements, tax forms, etc. are pervasive in business workflows. The tedious and error-prone nature of these workflows has led to much recent research into machine learning methods for automatically extracting structured information from such docum... | We first describe how a typical annotation task is set up to acquire labeled documents. We point out two major deficiencies with this approach before outlining an alternative that takes advantage of the characteristics of this domain. We then describe the assumptions underlying our approach. Given a document type for ... | 910 | 2,166 | 910 |
A Semi-Markov Structured Support Vector Machine Model for High-Precision Named Entity Recognition | Named entity recognition (NER) is the backbone of many NLP solutions. F 1 score, the harmonic mean of precision and recall, is often used to select/evaluate the best models. However, when precision needs to be prioritized over recall, a state-of-the-art model might not be the best choice. There is little in the literat... | Named Entity Recognition (NER) is the task of locating and categorizing phrases into a closed set of classes, such as organizations, people, and locations. NER is an information extraction task that is important for understanding large bodies of text and is an essential component for many natural language processing (N... | For classification, several papers try to optimize different evaluation metrics directly. Cost-sensitive classification For sequence tagging problems, inference-time heuristics for tuning the precision-recall trade-off for information extraction models have been pro- We adopt the BiLSTM-CNNs architecture embedding At t... | 1,031 | 1,295 | 1,031 |
Dealing with Semantic Underspecification in Multimodal NLP | Intelligent systems that aim at mastering language as humans do must deal with its semantic underspecification, namely, the possibility for a linguistic signal to convey only part of the information needed for communication to succeed. Consider the usages of the pronoun they, which can leave the gender and number of it... | They put the flowers there. Speakers of a language hear sentences like this every day and have no trouble understanding what they mean-and what message they convey. This is because, in a normal state of affairs, they can count on a wide range of information from the surrounding context, personal knowledge and experienc... | Semantic Underspecification? The field of multimodal or visually grounded NLP is currently dominated by pre-trained multimodal Transformers. Since their introduction, models like CLIP These models differ from each other in several dimensions. For example, they either concatenate and jointly process the visual and textu... | 1,399 | 2,023 | 1,399 |
NLP Workbench: Efficient and Extensible Integration of State-of-the-art Text Mining Tools | NLP Workbench is a web-based platform for text mining that allows non-expert users to obtain semantic understanding of large-scale corpora using state-of-the-art text mining models. The platform is built upon latest pretrained models and open source systems from academia that provide semantic analysis functionalities, ... | Text mining, also known as text analytics or text analysis, is the process where a user interacts with machine-supported analysis tools that transform natural language text into structured data, to gain insights and new knowledge from the text Nearly every subfield of NLP involved in text mining has been rapidly evolvi... | Figure The connected Neo4j Browser 11 provides an interactive web interface for exploring social networks constructed from a corpus. System Perspective From the perspective of the system, both the corpus and the outputs of text mining tools are stored and indexed in Elasticsearch prove efficiency. In addition to that, ... | 1,085 | 2,779 | 1,085 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.