The task of persona-steered text generation requires large language models (LLMs) to generate text that reflects the distribution of views that an individual fitting a persona could have. People have multifaceted personas, but prior work on bias in LLM-generated opinions has only explored multiple-choice settings or one-dimensional personas. We define an incongruous persona as a persona with multiple traits where one trait makes its other traits less likely in human survey data, e.g. political liberals who support increased military spending. We find that LLMs are 9.7% less steerable towards incongruous personas than congruous ones, sometimes generating the stereotypical stance associated with its demographic rather than the target stance. Models that we evaluate that are fine-tuned with Reinforcement Learning from Human Feedback (RLHF) are more steerable, especially towards stances associated with political liberals and women, but present significantly less diverse views of personas. We also find variance in LLM steerability that cannot be predicted from multiple-choice opinion evaluation. Our results show the importance of evaluating models in open-ended text generation, as it can surface new LLM opinion biases. Moreover, such a setup can shed light on our ability to steer models toward a richer and more diverse range of viewpoints.
Wavelet transforms, a powerful mathematical tool, have been widely used in different domains, including Signal and Image processing, to unravel intricate patterns, enhance data representation, and extract meaningful features from data. Tangible results from their application suggest that Wavelet transforms can be applied to NLP capturing a variety of linguistic and semantic properties.In this paper, we empirically leverage the application of Discrete Wavelet Transforms (DWT) to word and sentence embeddings. We aim to showcase the capabilities of DWT in analyzing embedding representations at different levels of resolution and compressing them while maintaining their overall quality.We assess the effectiveness of DWT embeddings on semantic similarity tasks to show how DWT can be used to consolidate important semantic information in an embedding vector. We show the efficacy of the proposed paradigm using different embedding models, including large language models, on downstream tasks. Our results show that DWT can reduce the dimensionality of embeddings by 50-93% with almost no change in performance for semantic similarity tasks, while achieving superior accuracy in most downstream tasks. Our findings pave the way for applying DWT to improve NLP applications.
In an era of model and data proliferation in machine learning/AI especially marked by the rapid advancement of open-sourced technologies, there arises a critical need for standardized consistent documentation. Our work addresses the information incompleteness in current human-written model and data cards. We propose an automated generation approach using Large Language Models (LLMs). Our key contributions include the establishment of CardBench, a comprehensive dataset aggregated from over 4.8k model cards and 1.4k data cards, coupled with the development of the CardGen pipeline comprising a two-step retrieval process. Our approach exhibits enhanced completeness, objectivity, and faithfulness in generated model and data cards, a significant step in responsible AI documentation practices ensuring better accountability and traceability.
Traditionally, natural language processing (NLP) models often use a rich set of features created by linguistic expertise, such as semantic representations. However, in the era of large language models (LLMs), more and more tasks are turned into generic, end-to-end sequence generation problems. In this paper, we investigate the question: what is the role of semantic representations in the era of LLMs? Specifically, we investigate the effect of Abstract Meaning Representation (AMR) across five diverse NLP tasks. We propose an AMR-driven chain-of-thought prompting method, which we call AMRCOT, and find that it generally hurts performance more than it helps. To investigate what AMR may have to offer on these tasks, we conduct a series of analysis experiments. We find that it is difficult to predict which input examples AMR may help or hurt on, but errors tend to arise with multi-word expressions, named entities, and in the final inference step where the LLM must connect its reasoning over the AMR to its prediction. We recommend focusing on these areas for future work in semantic representations for LLMs. Our code: https://github.com/causalNLP/amr_llm
The intricate relationship between language and culture has long been a subject of exploration within the realm of linguistic anthropology. Large Language Models (LLMs), promoted as repositories of collective human knowledge, raise a pivotal question: do these models genuinely encapsulate the diverse knowledge adopted by different cultures? Our study reveals that these models demonstrate greater cultural alignment along two dimensions—firstly, when prompted with the dominant language of a specific culture, and secondly, when pretrained with a refined mixture of languages employed by that culture. We quantify cultural alignment by simulating sociological surveys, comparing model responses to those of actual survey participants as references. Specifically, we replicate a survey conducted in various regions of Egypt and the United States through prompting LLMs with different pretraining data mixtures in both Arabic and English with the personas of the real respondents and the survey questions. Further analysis reveals that misalignment becomes more pronounced for underrepresented personas and for culturally sensitive topics, such as those probing social values. Finally, we introduce Anthropological Prompting, a novel method leveraging anthropological reasoning to enhance cultural alignment. Our study emphasizes the necessity for a more balanced multilingual pretraining dataset to better represent the diversity of human experience and the plurality of different cultures with many implications on the topic of cross-lingual transfer.
We present an overview of the FIGNEWSshared task, organized as part of the Arabic-NLP 2024 conference co-located with ACL2024. The shared task addresses bias and pro-paganda annotation in multilingual news posts.We focus on the early days of the Israel War onGaza as a case study. The task aims to fostercollaboration in developing annotation guide-lines for subjective tasks by creating frame-works for analyzing diverse narratives high-lighting potential bias and propaganda. In aspirit of fostering and encouraging diversity,we address the problem from a multilingualperspective, namely within five languages: En-glish, French, Arabic, Hebrew, and Hindi. Atotal of 17 teams participated in two annota-tion subtasks: bias (16 teams) and propaganda(6 teams). The teams competed in four evalua-tion tracks: guidelines development, annotationquality, annotation quantity, and consistency.Collectively, the teams produced 129,800 datapoints. Key findings and implications for thefield are discussed.
Language Models pretrained on large textual data have been shown to encode different types of knowledge simultaneously. Traditionally, only the features from the last layer are used when adapting to new tasks or data. We put forward that, when using or finetuning deep pretrained models, intermediate layer features that may be relevant to the downstream task are buried too deep to be used efficiently in terms of needed samples or steps. To test this, we propose a new layer fusion method: Depth-Wise Attention (DWAtt), to help re-surface signals from non-final layers. We compare DWAtt to a basic concatenation-based layer fusion method (Concat), and compare both to a deeper model baseline—all kept within a similar parameter budget. Our findings show that DWAtt and Concat are more step- and sample-efficient than the baseline, especially in the few-shot setting. DWAtt outperforms Concat on larger data sizes. On CoNLL-03 NER, layer fusion shows 3.68 − 9.73% F1 gain at different few-shot sizes. The layer fusion models presented significantly outperform the baseline in various training scenarios with different data sizes, architectures, and training constraints.
Recent advancements in large language models have enabled them to perform well on complex tasks that require step-by-step reasoning with few-shot learning. However, it is unclear whether these models are applying reasoning skills they have learnt during pre-training , or if they are simply memorizing their training corpus at finer granularity and have learnt to better understand their context. To address this question, we introduce {pasted macro ‘OUR’}model, a benchmark and suite of analyses for evaluating reasoning skills of language models. {pasted macro ‘OUR’}model enables comparing pre-trained and finetuned models on complex tasks that require reasoning skills to solve. Our benchmark provides a test bed to asses any language model on fine-grained reasoning skills, which spans over 20 datasets and covers 10 different reasoning skills. By using {pasted macro ‘OUR’}model we further investigate the role of finetuning. Our extensive empirical analysis shows that language models learn more reasoning skills such as textual entailment, abductive reasoning, and analogical reasoning during the finetuning stage compared to pretraining stage. However, we also find that when language models are finetuned they tend to overfit to the prompt template, which hurts the robustness of models causing generalization problems.
We present the ACL 60/60 evaluation sets for multilingual translation of ACL 2022 technical presentations into 10 target languages. This dataset enables further research into multilingual speech translation under realistic recording conditions with unsegmented audio and domain-specific terminology, applying NLP tools to text and speech in the technical domain, and evaluating and improving model robustness to diverse speaker demographics.
Language models can memorize a considerable amount of factual information during pretraining that can be elicited through prompting or finetuning models on tasks like question answering. In this paper, we discuss approaches to measuring model factual beliefs, updating incorrect factual beliefs in models, and visualizing graphical relationships between factual beliefs. Our main contributions include: (1) new metrics for evaluating belief-updating methods focusing on the logical consistency of beliefs, (2) a training objective for Sequential, Local, and Generalizing updates (SLAG) that improves the performance of existing hypernetwork approaches, and (3) the introduction of the belief graph, a new form of visualization for language models that shows relationships between stored model beliefs. Our experiments suggest that models show only limited consistency between factual beliefs, but update methods can both fix incorrect model beliefs and greatly improve their consistency. Although off-the-shelf optimizers are surprisingly strong belief-updating baselines, our learned optimizers can outperform them in more difficult settings than have been considered in past work.
Summarizing medical conversations is one of the tasks proposed by MEDIQA-Chat to promote research on automatic clinical note generation from doctor-patient conversations. In this paper, we present our submission to this task using fine-tuned language models, including T5, BART and BioGPT models. The fine-tuned models are evaluated using ensemble metrics including ROUGE, BERTScore andBLEURT. Among the fine-tuned models, Flan-T5 achieved the highest aggregated score for dialogue summarization.
Many NLP classification tasks, such as sexism/racism detection or toxicity detection, are based on human values. Yet, human values can vary under diverse cultural conditions. Therefore, we introduce a framework for value-aligned classification that performs prediction based on explicitly written human values in the command. Along with the task, we propose a practical approach that distills value-aligned knowledge from large-scale language models (LLMs) to construct value-aligned classifiers in two steps. First, we generate value-aligned training data from LLMs by prompt-based few-shot learning. Next, we fine-tune smaller classification models with the generated data for the task. Empirical results show that our VA-Models surpass multiple baselines by at least 15.56% on the F1-score, including few-shot learning with OPT-175B and existing text augmentation methods. We suggest that using classifiers with explicit human value input improves both inclusivity & explainability in AI.
We conduct a thorough investigation into the reasoning capabilities of Large Language Models (LLMs), focusing specifically on the Open Pretrained Transformers (OPT) models as a representative of such models. Our study entails finetuning three different sizes of OPT on a carefully curated reasoning corpus, resulting in two sets of finetuned models: OPT-R, finetuned without explanations, and OPT-RE, finetuned with explanations. We then evaluate all models on 57 out-of-domain tasks drawn from the Super-NaturalInstructions benchmark, covering 26 distinct reasoning skills, utilizing three prompting techniques. Through a comprehensive grid of 27 configurations and 6,156 test evaluations, we investigate the dimensions of finetuning, prompting, and scale to understand the role of explanations on different reasoning skills. Our findings reveal that having explanations in the fewshot exemplar has no significant impact on the model’s performance when the model is finetuned, while positively affecting the non-finetuned counterpart. Moreover, we observe a slight yet consistent increase in classification accuracy as we incorporate explanations during prompting and finetuning, respectively. Finally, we offer insights on which reasoning skills benefit the most from incorporating explanations during finetuning and prompting, such as Numerical (+20.4%) and Analogical (+13.9%) reasoning, as well as skills that exhibit negligible or negative effects.
Community Question Answering (CQA) fora such as Stack Overflow and Yahoo! Answers contain a rich resource of answers to a wide range of community-based questions. Each question thread can receive a large number of answers with different perspectives. One goal of answer summarization is to produce a summary that reflects the range of answer perspectives. A major obstacle for this task is the absence of a dataset to provide supervision for producing such summaries. Recent works propose heuristics to create such data, but these are often noisy and do not cover all answer perspectives present. This work introduces a novel dataset of 4,631 CQA threads for answer summarization curated by professional linguists. Our pipeline gathers annotations for all subtasks of answer summarization, including relevant answer sentence selection, grouping these sentences based on perspectives, summarizing each perspective, and producing an overall summary. We analyze and benchmark state-of-the-art models on these subtasks and introduce a novel unsupervised approach for multi-perspective data augmentation that boosts summarization performance according to automatic evaluation. Finally, we propose reinforcement learning rewards to improve factual consistency and answer coverage and analyze areas for improvement.
Obtaining meaningful quality scores for machine translation systems through human evaluation remains a challenge given the high variability between human evaluators, partly due to subjective expectations for translation quality for different language pairs. We propose a new metric called XSTS that is more focused on semantic equivalence and a cross-lingual calibration method that enables more consistent assessment. We demonstrate the effectiveness of these novel contributions in large scale evaluation studies across up to 14 language pairs, with translation both into and out of English.
Hate speech detection is complex; it relies on commonsense reasoning, knowledge of stereotypes, and an understanding of social nuance that differs from one culture to the next. It is also difficult to collect a large-scale hate speech annotated dataset. In this work, we frame this problem as a few-shot learning task, and show significant gains with decomposing the task into its “constituent” parts. In addition, we see that infusing knowledge from reasoning datasets (e.g. ATOMIC2020) improves the performance even further. Moreover, we observe that the trained models generalize to out-of-distribution datasets, showing the superiority of task decomposition and knowledge infusion compared to previously used methods. Concretely, our method outperforms the baseline by 17.83% absolute gain in the 16-shot case.
Large-scale generative language models such as GPT-3 are competitive few-shot learners. While these models are known to be able to jointly represent many different languages, their training data is dominated by English, potentially limiting their cross-lingual generalization. In this work, we train multilingual generative language models on a corpus covering a diverse set of languages, and study their few- and zero-shot learning capabilities in a wide range of tasks. Our largest model with 7.5 billion parameters sets new state of the art in few-shot learning in more than 20 representative languages, outperforming GPT-3 of comparable size in multilingual commonsense reasoning (with +7.4% absolute accuracy improvement in 0-shot settings and +9.4% in 4-shot settings) and natural language inference (+5.4% in each of 0-shot and 4-shot settings). On the FLORES-101 machine translation benchmark, our model outperforms GPT-3 on 171 out of 182 directions with 32 training examples, while surpassing the official supervised baseline in 45 directions. We conduct an in-depth analysis of different multilingual prompting approaches, showing in particular that strong few-shot learning performance across languages can be achieved via cross-lingual transfer through both templates and demonstration examples.
Mixture of Experts layers (MoEs) enable efficient scaling of language models through conditional computation. This paper presents a detailed empirical study of how autoregressive MoE language models scale in comparison with dense models in a wide range of settings: in- and out-of-domain language modeling, zero- and few-shot priming, and full-shot fine-tuning. With the exception of fine-tuning, we find MoEs to be substantially more compute efficient. At more modest training budgets, MoEs can match the performance of dense models using ~4 times less compute. This gap narrows at scale, but our largest MoE model (1.1T parameters) consistently outperforms a compute-equivalent dense model (6.7B parameters). Overall, this performance gap varies greatly across tasks and domains, suggesting that MoE and dense models generalize differently in ways that are worthy of future study. We make our code and models publicly available for research use.
When building NLP models, there is a tendency to aim for broader coverage, often overlooking cultural and (socio)linguistic nuance. In this position paper, we make the case for care and attention to such nuances, particularly in dataset annotation, as well as the inclusion of cultural and linguistic expertise in the process. We present a playbook for responsible dataset creation for polyglossic, multidialectal languages. This work is informed by a study on Arabic annotation of social media content.
In this paper, we tackle the Arabic Fine-Grained Hate Speech Detection shared task and demonstrate significant improvements over reported baselines for its three subtasks. The tasks are to predict if a tweet contains (1) Offensive language; and whether it is considered (2) Hate Speech or not and if so, then predict the (3) Fine-Grained Hate Speech label from one of six categories. Our final solution is an ensemble of models that employs multitask learning and a self-consistency correction method yielding 82.7% on the hate speech subtask—reflecting a 3.4% relative improvement compared to previous work.
We present the BeSt corpus, which records cognitive state: who believes what (i.e., factuality), and who has what sentiment towards what. This corpus is inspired by similar source-and-target corpora, specifically MPQA and FactBank. The corpus comprises two genres, newswire and discussion forums, in three languages, Chinese (Mandarin), English, and Spanish. The corpus is distributed through the LDC.
We present a tool, Text Characterization Toolkit (TCT), that researchers can use to study characteristics of large datasets. Furthermore, such properties can lead to understanding the influence of such attributes on models’ behaviour. Traditionally, in most NLP research, models are usually evaluated by reporting single-number performance scores on a number of readily available benchmarks, without much deeper analysis. Here, we argue that – especially given the well-known fact that benchmarks often contain biases, artefacts, and spurious correlations – deeper results analysis should become the de-facto standard when presenting new models or benchmarks. TCT aims at filling this gap by facilitating such deeper analysis for datasets at scale, where datasets can be for training/development/evaluation. TCT includes both an easy-to-use tool, as well as off-the-shelf scripts that can be used for specific analyses. We also present use-cases from several different domains. TCT is used to predict difficult examples for given well-known trained models; TCT is also used to identify (potentially harmful) biases present in a dataset.
Decision making theories such as Fuzzy-Trace Theory (FTT) suggest that individuals tend to rely on gist, or bottom-line meaning, in the text when making decisions. In this work, we delineate the process of developing GisPy, an opensource tool in Python for measuring the Gist Inference Score (GIS) in text. Evaluation of GisPy on documents in three benchmarks from the news and scientific text domains demonstrates that scores generated by our tool significantly distinguish low vs. high gist documents. Our tool is publicly available to use at: https: //github.com/phosseini/GisPy.
Schizophrenia is one of the most disabling mental health conditions to live with. Approximately one percent of the population has schizophrenia which makes it fairly common, and it affects many people and their families. Patients with schizophrenia suffer different symptoms: formal thought disorder (FTD), delusions, and emotional flatness. In this paper, we quantitatively and qualitatively analyze the language of patients with schizophrenia measuring various linguistic features in two modalities: speech and written text. We examine the following features: coherence and cohesion of thoughts, emotions, specificity, level of commit- ted belief (LCB), and personality traits. Our results show that patients with schizophrenia score high in fear and neuroticism compared to healthy controls. In addition, they are more committed to their beliefs, and their writing lacks details. They score lower in most of the linguistic features of cohesion with significant p-values.
Previous studies have shown the efficacy of knowledge augmentation methods in pretrained language models. However, these methods behave differently across domains and downstream tasks. In this work, we investigate the augmentation of pretrained language models with knowledge graph data in the cause-effect relation classification and commonsense causal reasoning tasks. After automatically verbalizing triples in ATOMIC2020, a wide coverage commonsense reasoning knowledge graph, we continually pretrain BERT and evaluate the resulting model on cause-effect pair classification and answering commonsense causal reasoning questions. Our results show that a continually pretrained language model augmented with commonsense reasoning knowledge outperforms our baselines on two commonsense causal reasoning benchmarks, COPA and BCOPA-CE, and a Temporal and Causal Reasoning (TCR) dataset, without additional improvement in model architecture or using quality-enhanced data for fine-tuning.
The scarcity of parallel data is a major obstacle for training high-quality machine translation systems for low-resource languages. Fortunately, some low-resource languages are linguistically related or similar to high-resource languages; these related languages may share many lexical or syntactic structures. In this work, we exploit this linguistic overlap to facilitate translating to and from a low-resource language with only monolingual data, in addition to any parallel data in the related high-resource language. Our method, NMT-Adapt, combines denoising autoencoding, back-translation and adversarial objectives to utilize monolingual data for low-resource adaptation. We experiment on 7 languages from three different language families and show that our technique significantly improves translation into low-resource language compared to other translation baselines.
Is bias amplified when neural machine translation (NMT) models are optimized for speed and evaluated on generic test sets using BLEU? We investigate architectures and techniques commonly used to speed up decoding in Transformer-based models, such as greedy search, quantization, average attention networks (AANs) and shallow decoder models and show their effect on gendered noun translation. We construct a new gender bias test set, SimpleGEN, based on gendered noun phrases in which there is a single, unambiguous, correct answer. While we find minimal overall BLEU degradation as we apply speed optimizations, we observe that gendered noun translation performance degrades at a much faster rate.
Modern sentence encoders are used to generate dense vector representations that capture the underlying linguistic characteristics for a sequence of words, including phrases, sentences, or paragraphs. These kinds of representations are ideal for training a classifier for an end task such as sentiment analysis, question answering and text classification. Different models have been proposed to efficiently generate general purpose sentence representations to be used in pretraining protocols. While averaging is the most commonly used efficient sentence encoder, Discrete Cosine Transform (DCT) was recently proposed as an alternative that captures the underlying syntactic characteristics of a given text without compromising practical efficiency compared to averaging. However, as with most other sentence encoders, the DCT sentence encoder was only evaluated in English. To this end, we utilize DCT encoder to generate universal sentence representation for different languages such as German, French, Spanish and Russian. The experimental results clearly show the superior effectiveness of DCT encoding in which consistent performance improvements are achieved over strong baselines on multiple standardized datasets
Social media has emerged as a key channel for seeking information. Online users spend several hours reading, posting, and searching for news on microblogging platforms daily. However, this could act as a double-edged sword especially when not all information online is reliable. Moreover, the inherently unmoderated nature of social media renders identifying unverified information ever more challenging. Most of the existing approaches for rumor tracking are not scalable because of their dependency on a significant amount of labeled data. In this work, we investigate this problem from different angles. We design an Active-Transfer Learning (ATL) strategy to identify rumors with a limited amount of annotated data. We go beyond that and investigate the impact of leveraging various machine learning approaches in addition to different contextual representations. We discuss the impact of multiple classifiers on a limited amount of annotated data followed by an interactive approach to gradually update the models by adding the least certain samples (LCS) from the pool of unlabeled data. Our proposed Active Learning (AL) strategy achieves faster convergence in terms of the F-score while requiring fewer annotated samples (42% of the whole dataset for the best model).
We release an urgency dataset that consists of English tweets relating to natural crises, along with annotations of their corresponding urgency status. Additionally, we release evaluation datasets for two low-resource languages, i.e. Sinhala and Odia, and demonstrate an effective zero-shot transfer from English to these two languages by training cross-lingual classifiers. We adopt cross-lingual embeddings constructed using different methods to extract features of the tweets, including a few state-of-the-art contextual embeddings such as BERT, RoBERTa and XLM-R. We train classifiers of different architectures on the extracted features. We also explore semi-supervised approaches by utilizing unlabeled tweets and experiment with ensembling different classifiers. With very limited amounts of labeled data in English and zero data in the low resource languages, we show a successful framework of training monolingual and cross-lingual classifiers using deep learning methods which are known to be data hungry. Specifically, we show that the recent deep contextual embeddings are also helpful when dealing with very small-scale datasets. Classifiers that incorporate RoBERTa yield the best performance for English urgency detection task, with F1 scores that are more than 25 points over our baseline classifier. For the zero-shot transfer to low resource languages, classifiers that use LASER features perform the best for Sinhala transfer while XLM-R features benefit the Odia transfer the most.
Summarizing data samples by quantitative measures has a long history, with descriptive statistics being a case in point. However, as natural language processing methods flourish, there are still insufficient characteristic metrics to describe a collection of texts in terms of the words, sentences, or paragraphs they comprise. In this work, we propose metrics of diversity, density, and homogeneity that quantitatively measure the dispersion, sparsity, and uniformity of a text collection. We conduct a series of simulations to verify that each metric holds desired properties and resonates with human intuitions. Experiments on real-world datasets demonstrate that the proposed characteristic metrics are highly correlated with text classification performance of a renowned model, BERT, which could inspire future applications.
Neural abstractive summarization models are prone to generate content inconsistent with the source document, i.e. unfaithful. Existing automatic metrics do not capture such mistakes effectively. We tackle the problem of evaluating faithfulness of a generated summary given its source document. We first collected human annotations of faithfulness for outputs from numerous models on two datasets. We find that current models exhibit a trade-off between abstractiveness and faithfulness: outputs with less word overlap with the source document are more likely to be unfaithful. Next, we propose an automatic question answering (QA) based metric for faithfulness, FEQA, which leverages recent advances in reading comprehension. Given question-answer pairs generated from the summary, a QA model extracts answers from the document; non-matched answers indicate unfaithful information in the summary. Among metrics based on word overlap, embedding similarity, and learned language understanding models, our QA-based metric has significantly higher correlation with human faithfulness scores, especially on highly abstractive summaries.
In many languages like Arabic, diacritics are used to specify pronunciations as well as meanings. Such diacritics are often omitted in written text, increasing the number of possible pronunciations and meanings for a word. This results in a more ambiguous text making computational processing on such text more difficult. Diacritic restoration is the task of restoring missing diacritics in the written text. Most state-of-the-art diacritic restoration models are built on character level information which helps generalize the model to unseen data, but presumably lose useful information at the word level. Thus, to compensate for this loss, we investigate the use of multi-task learning to jointly optimize diacritic restoration with related NLP problems namely word segmentation, part-of-speech tagging, and syntactic diacritization. We use Arabic as a case study since it has sufficient data resources for tasks that we consider in our joint modeling. Our joint models significantly outperform the baselines and are comparable to the state-of-the-art models that are more complex relying on morphological analyzers and/or a lot more data (e.g. dialectal data).
The increased focus on misinformation has spurred development of data and systems for detecting the veracity of a claim as well as retrieving authoritative evidence. The Fact Extraction and VERification (FEVER) dataset provides such a resource for evaluating endto- end fact-checking, requiring retrieval of evidence from Wikipedia to validate a veracity prediction. We show that current systems for FEVER are vulnerable to three categories of realistic challenges for fact-checking – multiple propositions, temporal reasoning, and ambiguity and lexical variation – and introduce a resource with these types of claims. Then we present a system designed to be resilient to these “attacks” using multiple pointer networks for document selection and jointly modeling a sequence of evidence sentences and veracity relation predictions. We find that in handling these attacks we obtain state-of-the-art results on FEVER, largely due to improved evidence retrieval.
Intent classification (IC) and slot filling (SF) are core components in most goal-oriented dialogue systems. Current IC/SF models perform poorly when the number of training examples per class is small. We propose a new few-shot learning task, few-shot IC/SF, to study and improve the performance of IC and SF models on classes not seen at training time in ultra low resource scenarios. We establish a few-shot IC/SF benchmark by defining few-shot splits for three public IC/SF datasets, ATIS, TOP, and Snips. We show that two popular few-shot learning algorithms, model agnostic meta learning (MAML) and prototypical networks, outperform a fine-tuning baseline on this benchmark. Prototypical networks achieves significant gains in IC performance on the ATIS and TOP datasets, while both prototypical networks and MAML outperform the baseline with respect to SF on all three datasets. In addition, we demonstrate that joint training as well as the use of pre-trained language models, ELMo and BERT in our case, are complementary to these few-shot learning methods and yield further gains.
We describe a method for developing broad-coverage semantic dependency parsers for languages for which no semantically annotated resource is available. We leverage a multitask learning framework coupled with annotation projection. We use syntactic parsing as the auxiliary task in our multitask setup. Our annotation projection experiments from English to Czech show that our multitask setup yields 3.1% (4.2%) improvement in labeled F1-score on in-domain (out-of-domain) test set compared to a single-task baseline.
We develop and investigate several cross-lingual alignment approaches for neural sentence embedding models, such as the supervised inference classifier, InferSent, and sequential encoder-decoder models. We evaluate three alignment frameworks applied to these models: joint modeling, representation transfer learning, and sentence mapping, using parallel text to guide the alignment. Our results support representation transfer as a scalable approach for modular cross-lingual alignment of neural sentence embeddings, where we observe better performance compared to joint models in intrinsic and extrinsic evaluations, particularly with smaller sets of parallel data.
In this paper we present an emotion classifier models that submitted to the SemEval-2019 Task 3 : EmoContext. Our approach is a Gated Recurrent Neural Network (GRU) model with attention layer is bootstrapped with contextual information and trained with a multigenre corpus, which is combination of several popular emotional data sets. We utilize different word embeddings to empirically select the most suited embedding to represent our features. Our aim is to build a robust emotion classifier that can generalize emotion detection, which is to learn emotion cues in a noisy training environment. To fulfill this aim we train our model with a multigenre emotion corpus, this way we leverage from having more training set. We achieved overall %56.05 f1-score and placed 144. Given our aim and noisy training environment, the results are anticipated.
Social media plays a crucial role as the main resource news for information seekers online. However, the unmoderated feature of social media platforms lead to the emergence and spread of untrustworthy contents which harm individuals or even societies. Most of the current automated approaches for automatically determining the veracity of a rumor are not generalizable for novel emerging topics. This paper describes our hybrid system comprising rules and a machine learning model which makes use of replied tweets to identify the veracity of the source tweet. The proposed system in this paper achieved 0.435 F-Macro in stance classification, and 0.262 F-macro and 0.801 RMSE in rumor verification tasks in Task7 of SemEval 2019.
Cross-lingual word vectors are typically obtained by fitting an orthogonal matrix that maps the entries of a bilingual dictionary from a source to a target vector space. Word vectors, however, are most commonly used for sentence or document-level representations that are calculated as the weighted average of word embeddings. In this paper, we propose an alternative to word-level mapping that better reflects sentence-level cross-lingual similarity. We incorporate context in the transformation matrix by directly mapping the averaged embeddings of aligned sentences in a parallel corpus. We also implement cross-lingual mapping of deep contextualized word embeddings using parallel sentences with word alignments. In our experiments, both approaches resulted in cross-lingual sentence embeddings that outperformed context-independent word mapping in sentence translation retrieval. Furthermore, the sentence-level transformation could be used for word-level mapping without loss in word translation quality.
Natural Language Understanding (NLU) is a core component of dialog systems. It typically involves two tasks - Intent Classification (IC) and Slot Labeling (SL), which are then followed by a dialogue management (DM) component. Such NLU systems cater to utterances in isolation, thus pushing the problem of context management to DM. However, contextual information is critical to the correct prediction of intents in a conversation. Prior work on contextual NLU has been limited in terms of the types of contextual signals used and the understanding of their impact on the model. In this work, we propose a context-aware self-attentive NLU (CASA-NLU) model that uses multiple signals over a variable context window, such as previous intents, slots, dialog acts and utterances, in addition to the current user utterance. CASA-NLU outperforms a recurrent contextual NLU baseline on two conversational datasets, yielding a gain of up to 7% on the IC task. Moreover, a non-contextual variant of CASA-NLU achieves state-of-the-art performance on standard public datasets - SNIPS and ATIS.
Diacritic restoration has gained importance with the growing need for machines to understand written texts. The task is typically modeled as a sequence labeling problem and currently Bidirectional Long Short Term Memory (BiLSTM) models provide state-of-the-art results. Recently, Bai et al. (2018) show the advantages of Temporal Convolutional Neural Networks (TCN) over Recurrent Neural Networks (RNN) for sequence modeling in terms of performance and computational resources. As diacritic restoration benefits from both previous as well as subsequent timesteps, we further apply and evaluate a variant of TCN, Acausal TCN (A-TCN), which incorporates context from both directions (previous and future) rather than strictly incorporating previous context as in the case of TCN. A-TCN yields significant improvement over TCN for diacritization in three different languages: Arabic, Yoruba, and Vietnamese. Furthermore, A-TCN and BiLSTM have comparable performance, making A-TCN an efficient alternative over BiLSTM since convolutions can be trained in parallel. A-TCN is significantly faster than BiLSTM at inference time (270% 334% improvement in the amount of text diacritized per minute).
Vector averaging remains one of the most popular sentence embedding methods in spite of its obvious disregard for syntactic structure. While more complex sequential or convolutional networks potentially yield superior classification performance, the improvements in classification accuracy are typically mediocre compared to the simple vector averaging. As an efficient alternative, we propose the use of discrete cosine transform (DCT) to compress word sequences in an order-preserving manner. The lower order DCT coefficients represent the overall feature patterns in sentences, which results in suitable embeddings for tasks that could benefit from syntactic features. Our results in semantic probing tasks demonstrate that DCT embeddings indeed preserve more syntactic information compared with vector averaging. With practically equivalent complexity, the model yields better overall performance in downstream classification tasks that correlate with syntactic features, which illustrates the capacity of DCT to preserve word order information.
The need for high-quality, large-scale, goal-oriented dialogue datasets continues to grow as virtual assistants become increasingly wide-spread. However, publicly available datasets useful for this area are limited either in their size, linguistic diversity, domain coverage, or annotation granularity. In this paper, we present strategies toward curating and annotating large scale goal oriented dialogue data. We introduce the MultiDoGO dataset to overcome these limitations. With a total of over 81K dialogues harvested across six domains, MultiDoGO is over 8 times the size of MultiWOZ, the other largest comparable dialogue dataset currently available to the public. Over 54K of these harvested conversations are annotated for intent classes and slot labels. We adopt a Wizard-of-Oz approach wherein a crowd-sourced worker (the “customer”) is paired with a trained annotator (the “agent”). The data curation process was controlled via biases to ensure a diversity in dialogue flows following variable dialogue policies. We provide distinct class label tags for agents vs. customer utterances, along with applicable slot labels. We also compare and contrast our strategies on annotation granularity, i.e. turn vs. sentence level. Furthermore, we compare and contrast annotations curated by leveraging professional annotators vs the crowd. We believe our strategies for eliciting and annotating such a dialogue dataset scales across modalities and domains and potentially languages in the future. To demonstrate the efficacy of our devised strategies we establish neural baselines for classification on the agent and customer utterances as well as slot labeling for each domain.
The blurry line between nefarious fake news and protected-speech satire has been a notorious struggle for social media platforms. Further to the efforts of reducing exposure to misinformation on social media, purveyors of fake news have begun to masquerade as satire sites to avoid being demoted. In this work, we address the challenge of automatically classifying fake news versus satire. Previous work have studied whether fake news and satire can be distinguished based on language differences. Contrary to fake news, satire stories are usually humorous and carry some political or social message. We hypothesize that these nuances could be identified using semantic and linguistic cues. Consequently, we train a machine learning method using semantic representation, with a state-of-the-art contextual language model, and with linguistic features based on textual coherence metrics. Empirical evaluation attests to the merits of our approach compared to the language-based baseline and sheds light on the nuances between fake news and satire. As avenues for future work, we consider studying additional linguistic features related to the humor aspect, and enriching the data with current news events, to help identify a political or social message.
We describe a transfer method based on annotation projection to develop a dependency-based semantic role labeling system for languages for which no supervised linguistic information other than parallel data is available. Unlike previous work that presumes the availability of supervised features such as lemmas, part-of-speech tags, and dependency parse trees, we only make use of word and character features. Our deep model considers using character-based representations as well as unsupervised stem embeddings to alleviate the need for supervised features. Our experiments outperform a state-of-the-art method that uses supervised lexico-syntactic features on 6 out of 7 languages in the Universal Proposition Bank.
Linguistic Code Switching (CS) is a phenomenon that occurs when multilingual speakers alternate between two or more languages/dialects within a single conversation. Processing CS data is especially challenging in intra-sentential data given state-of-the-art monolingual NLP technologies since such technologies are geared toward the processing of one language at a time. In this paper, we address the problem of Part-of-Speech tagging (POS) in the context of linguistic code switching (CS). We explore leveraging multiple neural network architectures to measure the impact of different pre-trained embeddings methods on POS tagging CS data. We investigate the landscape in four CS language pairs, Spanish-English, Hindi-English, Modern Standard Arabic- Egyptian Arabic dialect (MSA-EGY), and Modern Standard Arabic- Levantine Arabic dialect (MSA-LEV). Our results show that multilingual embedding (e.g., MSA-EGY and MSA-LEV) helps closely related languages (EGY/LEV) but adds noise to the languages that are distant (SPA/HIN). Finally, we show that our proposed models outperform state-of-the-art CS taggers for MSA-EGY language pair.
Lexical ambiguity, a challenging phenomenon in all natural languages, is particularly prevalent for languages with diacritics that tend to be omitted in writing, such as Arabic. Omitting diacritics leads to an increase in the number of homographs: different words with the same spelling. Diacritic restoration could theoretically help disambiguate these words, but in practice, the increase in overall sparsity leads to performance degradation in NLP applications. In this paper, we propose approaches for automatically marking a subset of words for diacritic restoration, which leads to selective homograph disambiguation. Compared to full or no diacritic restoration, these approaches yield selectively-diacritized datasets that balance sparsity and lexical disambiguation. We evaluate the various selection strategies extrinsically on several downstream applications: neural machine translation, part-of-speech tagging, and semantic textual similarity. Our experiments on Arabic show promising results, where our devised strategies on selective diacritization lead to a more balanced and consistent performance in downstream applications.
We evaluated various compositional models, from bag-of-words representations to compositional RNN-based models, on several extrinsic supervised and unsupervised evaluation benchmarks. Our results confirm that weighted vector averaging can outperform context-sensitive models in most benchmarks, but structural features encoded in RNN models can also be useful in certain classification tasks. We analyzed some of the evaluation datasets to identify the aspects of meaning they measure and the characteristics of the various models that explain their performance variance.
Detection and classification of emotion categories expressed by a sentence is a challenging task due to subjectivity of emotion. To date, most of the models are trained and evaluated on single genre and when used to predict emotion in different genre their performance drops by a large margin. To address the issue of robustness, we model the problem within a joint multi-task learning framework. We train this model with a multigenre emotion corpus to predict emotions across various genre. Each genre is represented as a separate task, we use soft parameter shared layers across the various tasks. our experimental results show that this model improves the results across the various genres, compared to a single genre training in the same neural net architecture.
Most existing methods for automatic bilingual dictionary induction rely on prior alignments between the source and target languages, such as parallel corpora or seed dictionaries. For many language pairs, such supervised alignments are not readily available. We propose an unsupervised approach for learning a bilingual dictionary for a pair of languages given their independently-learned monolingual word embeddings. The proposed method exploits local and global structures in monolingual vector spaces to align them such that similar words are mapped to each other. We show empirically that the performance of bilingual correspondents that are learned using our proposed unsupervised method is comparable to that of using supervised bilingual correspondents from a seed dictionary.
In the third shared task of the Computational Approaches to Linguistic Code-Switching (CALCS) workshop, we focus on Named Entity Recognition (NER) on code-switched social-media data. We divide the shared task into two competitions based on the English-Spanish (ENG-SPA) and Modern Standard Arabic-Egyptian (MSA-EGY) language pairs. We use Twitter data and 9 entity types to establish a new dataset for code-switched NER benchmarks. In addition to the CS phenomenon, the diversity of the entities and the social media challenges make the task considerably hard to process. As a result, the best scores of the competitions are 63.76% and 71.61% for ENG-SPA and MSA-EGY, respectively. We present the scores of 9 participants and discuss the most common challenges among submissions.
Many tasks such as question answering and reading comprehension rely on information extracted from unreliable sources. These systems would thus benefit from knowing whether a statement from an unreliable source is correct. We present experiments on the FEVER (Fact Extraction and VERification) task, a shared task that involves selecting sentences from Wikipedia and predicting whether a claim is supported by those sentences, refuted, or there is not enough information. Fact checking is a task that benefits from not only asserting or disputing the veracity of a claim but also finding evidence for that position. As these tasks are dependent on each other, an ideal model would consider the veracity of the claim when finding evidence and also find only the evidence that is relevant. We thus jointly model sentence extraction and verification on the FEVER shared task. Among all participants, we ranked 5th on the blind test set (prior to any additional human evaluation of the evidence).
Schizophrenia is one of the most disabling and difficult to treat of all human medical/health conditions, ranking in the top ten causes of disability worldwide. It has been a puzzle in part due to difficulty in identifying its basic, fundamental components. Several studies have shown that some manifestations of schizophrenia (e.g., the negative symptoms that include blunting of speech prosody, as well as the disorganization symptoms that lead to disordered language) can be understood from the perspective of linguistics. However, schizophrenia research has not kept pace with technologies in computational linguistics, especially in semantics and pragmatics. As such, we examine the writings of schizophrenia patients analyzing their syntax, semantics and pragmatics. In addition, we analyze tweets of (self proclaimed) schizophrenia patients who publicly discuss their diagnoses. For writing samples dataset, syntactic features are found to be the most successful in classification whereas for the less structured Twitter dataset, a combination of features performed the best.
Semantic Textual Similarity (STS) measures the meaning similarity of sentences. Applications include machine translation (MT), summarization, generation, question answering (QA), short answer grading, semantic search, dialog and conversational systems. The STS shared task is a venue for assessing the current state-of-the-art. The 2017 task focuses on multilingual and cross-lingual pairs with one sub-track exploring MT quality estimation (MTQE) data. The task obtained strong participation from 31 teams, with 17 participating in all language tracks. We summarize performance and review a selection of well performing methods. Analysis highlights common errors, providing insight into the limitations of existing models. To support ongoing work on semantic representations, the STS Benchmark is introduced as a new shared training and evaluation set carefully selected from the corpus of English STS shared task data (2012-2017).
This paper describes our submission to SemEval-2017 Task 3 Subtask D, “Question Answer Ranking in Arabic Community Question Answering”. In this work, we applied a supervised machine learning approach to automatically re-rank a set of QA pairs according to their relevance to a given question. We employ features based on latent semantic models, namely WTMF, as well as a set of lexical features based on string lengths and surface level matching. The proposed system ranked first out of 3 submissions, with a MAP score of 61.16%.
Our paper addresses the problem of annotation projection for semantic role labeling for resource-poor languages using supervised annotations from a resource-rich language through parallel data. We propose a transfer method that employs information from source and target syntactic dependencies as well as word alignment density to improve the quality of an iterative bootstrapping method. Our experiments yield a 3.5 absolute labeled F-score improvement over a standard annotation projection method.
In this paper we present a system for automatic Arabic text diacritization using three levels of analysis granularity in a layered back off manner. We build and exploit diacritized language models (LM) for each of three different levels of granularity: surface form, morphologically segmented into prefix/stem/suffix, and character level. For each of the passes, we use Viterbi search to pick the most probable diacritization per word in the input. We start with the surface form LM, followed by the morphological level, then finally we leverage the character level LM. Our system outperforms all of the published systems evaluated against the same training and test data. It achieves a 10.87% WER for complete full diacritization including lexical and syntactic diacritization, and 3.0% WER for lexical diacritization, ignoring syntactic diacritization.
Determining the textual entailment between texts is important in many NLP tasks, such as summarization, question answering, and information extraction and retrieval. Various methods have been suggested based on external knowledge sources; however, such resources are not always available in all languages and their acquisition is typically laborious and very costly. Distributional word representations such as word embeddings learned over large corpora have been shown to capture syntactic and semantic word relationships. Such models have contributed to improving the performance of several NLP tasks. In this paper, we address the problem of textual entailment in Arabic. We employ both traditional features and distributional representations. Crucially, we do not depend on any external resources in the process. Our suggested approach yields state of the art performance on a standard data set, ArbTE, achieving an accuracy of 76.2 % compared to state of the art of 69.3 %.
Most diacritics in Arabic represent short vowels. In Arabic orthography, such diacritics are considered optional. The absence of these diacritics naturally leads to significant word ambiguity to top the inherent ambiguity present in fully diacritized words. Word ambiguity is a significant impediment for machine translation. Despite the ambiguity presented by lack of diacritization, context helps ameliorate the situation. Identifying the appropriate amount of diacritic restoration to reduce word sense ambiguity in the context of machine translation is the object of this paper. Diacritic marks help reduce the number of possible lexical word choices assigned to a source word which leads to better quality translated sentences. We investigate a variety of (linguistically motivated) partial diacritization schemes that preserve some of the semantics that in essence complement the implicit contextual information present in the sentences. We also study the effect of training data size and report results on three standard test sets that represent a combination of different genres. The results show statistically significant improvements for some schemes compared to two baselines: text with no diacritics (the typical writing system adopted for Arabic) and text that is fully diacritized.
Arabic writing is typically underspecified for short vowels and other markups, referred to as diacritics. In addition to the lexical ambiguity exhibited in most languages, the lack of diacritics in written Arabic adds another layer of ambiguity which is an artifact of the orthography. In this paper, we present the details of three annotation experimental conditions designed to study the impact of automatic ambiguity detection, on annotation speed and quality in a large scale annotation project.
This paper describes the GW/LT3 contribution to the 2016 VarDial shared task on the identification of similar languages (task 1) and Arabic dialects (task 2). For both tasks, we experimented with Logistic Regression and Neural Network classifiers in isolation. Additionally, we implemented a cascaded classifier that consists of coarse and fine-grained classifiers (task 1) and a classifier ensemble with majority voting for task 2. The submitted systems obtained state-of-the art performance and ranked first for the evaluation on social media data (test sets B1 and B2 for task 1), with a maximum weighted F1 score of 91.94%.
We recently witnessed an exponential growth in dialectal Arabic usage in both textual data and speech recordings especially in social media. Processing such media is of great utility for all kinds of applications ranging from information extraction to social media analytics for political and commercial purposes to building decision support systems. Compared to other languages, Arabic, especially the informal variety, poses a significant challenge to natural language processing algorithms since it comprises multiple dialects, linguistic code switching, and a lack of standardized orthographies, to top its relatively complex morphology. Inherently, the problem of processing Arabic in the context of social media is the problem of how to handle resource poor languages. In this talk I will go over some of our insights to some of these problems and show how there is a silver lining where we can generalize some of our solutions to other low resource language contexts.
We present an approach for automatic verification and augmentation of multilingual lexica. We exploit existing parallel and monolingual corpora to extract multilingual correspondents via tri-angulation. We demonstrate the efficacy of our approach on two publicly available resources: Tharwa, a three-way lexicon comprising Dialectal Arabic, Modern Standard Arabic and English lemmas among other information (Diab et al., 2014); and BabelNet, a multilingual thesaurus comprising over 276 languages including Arabic variant entries (Navigli and Ponzetto, 2012). Our automated approach yields an F1-score of 71.71% in generating correct multilingual correspondents against gold Tharwa, and 54.46% against gold BabelNet without any human intervention.
The interaction between roots and patterns in Arabic has intrigued lexicographers and morphologists for centuries. While roots provide the consonantal building blocks, patterns provide the syllabic vocalic moulds. While roots provide abstract semantic classes, patterns realize these classes in specific instances. In this way both roots and patterns are indispensable for understanding the derivational, morphological and, to some extent, the cognitive aspects of the Arabic language. In this paper we perform lemmatization (a high-level lexical processing) without relying on a lookup dictionary. We use a hybrid approach that consists of a machine learning classifier to predict the lemma pattern for a given stem, and mapping rules to convert stems to their respective lemmas with the vocalization defined by the pattern.
Although MWE are relatively morphologically and syntactically fixed expressions, several types of flexibility can be observed in MWE, verbal MWE in particular. Identifying the degree of morphological and syntactic flexibility of MWE is very important for many Lexicographic and NLP tasks. Adding MWE variants/tokens to a dictionary resource requires characterizing the flexibility among other morphosyntactic features. Carrying out the task manually faces several challenges since it is a very laborious task time and effort wise, as well as it will suffer from coverage limitation. The problem is exacerbated in rich morphological languages where the average word in Arabic could have 12 possible inflection forms. Accordingly, in this paper we introduce a semi-automatic Arabic multiwords expressions resource (SAMER). We propose an automated method that identifies the morphological and syntactic flexibility of Arabic Verbal Multiword Expressions (AVMWE). All observed morphological variants and syntactic pattern alternations of an AVMWE are automatically acquired using large scale corpora. We look for three morphosyntactic aspects of AVMWE types investigating derivational and inflectional variations and syntactic templates, namely: 1) inflectional variation (inflectional paradigm) and calculating degree of flexibility; 2) derivational productivity; and 3) identifying and classifying the different syntactic types. We build a comprehensive list of AVMWE. Every token in the AVMWE list is lemmatized and tagged with POS information. We then search Arabic Gigaword and All ATBs for all possible flexible matches. For each AVMWE type we generate: a) a statistically ranked list of MWE-lexeme inflections and syntactic pattern alternations; b) An abstract syntactic template; and c) The most frequent form. Our technique is validated using a Golden MWE annotated list. The results shows that the quality of the generated resource is 80.04%.
Idafa in traditional Arabic grammar is an umbrella construction that covers several phenomena including what is expressed in English as noun-noun compounds and Saxon and Norman genitives. Additionally, Idafa participates in some other constructions, such as quantifiers, quasi-prepositions, and adjectives. Identifying the various types of the Idafa construction (IC) is of importance to Natural Language processing (NLP) applications. Noun-Noun compounds exhibit special behavior in most languages impacting their semantic interpretation. Hence distinguishing them could have an impact on downstream NLP applications. The most comprehensive syntactic representation of the Arabic language is the LDC Arabic Treebank (ATB). In the ATB, ICs are not explicitly labeled and furthermore, there is no distinction between ICs of noun-noun relations and other traditional ICs. Hence, we devise a detailed syntactic and semantic typification process of the IC phenomenon in Arabic. We target the ATB as a platform for this classification. We render the ATB annotated with explicit IC labels but with the further semantic characterization which is useful for syntactic, semantic and cross language processing. Our typification of IC comprises 3 main syntactic IC types: FIC, GIC, and TIC, and they are further divided into 10 syntactic subclasses. The TIC group is further classified into semantic relations. We devise a method for automatic IC labeling and compare its yield against the CATiB treebank. Our evaluation shows that we achieve the same level of accuracy, but with the additional fine-grained classification into the various syntactic and semantic types.
This paper presents the annotation guidelines developed as part of an effort to create a large scale manually diacritized corpus for various Arabic text genres. The target size of the annotated corpus is 2 million words. We summarize the guidelines and describe issues encountered during the training of the annotators. We also discuss the challenges posed by the complexity of the Arabic language and how they are addressed. Finally, we present the diacritization annotation procedure and detail the quality of the resulting annotations.
Text preprocessing is an important and necessary task for all NLP applications. A simple variation in any preprocessing step may drastically affect the final results. Moreover replicability and comparability, as much as feasible, is one of the goals of our scientific enterprise, thus building systems that can ensure the consistency in our various pipelines would contribute significantly to our goals. The problem has become quite pronounced with the abundance of NLP tools becoming more and more available yet with different levels of specifications. In this paper, we present a dynamic unified preprocessing framework and tool, SPLIT, that is highly configurable based on user requirements which serves as a preprocessing tool for several tools at once. SPLIT aims to standardize the implementations of the most important preprocessing steps by allowing for a unified API that could be exchanged across different researchers to ensure complete transparency in replication. The user is able to select the required preprocessing tasks among a long list of preprocessing steps. The user is also able to specify the order of execution which in turn affects the final preprocessing output.
We present our effort to create a large Multi-Layered representational repository of Linguistic Code-Switched Arabic data. The process involves developing clear annotation standards and Guidelines, streamlining the annotation process, and implementing quality control measures. We used two main protocols for annotation: in-lab gold annotations and crowd sourcing annotations. We developed a web-based annotation tool to facilitate the management of the annotation process. The current version of the repository contains a total of 886,252 tokens that are tagged into one of sixteen code-switching tags. The data exhibits code switching between Modern Standard Arabic and Egyptian Dialectal Arabic representing three data genres: Tweets, commentaries, and discussion fora. The overall Inter-Annotator Agreement is 93.1%.
We introduce a generic Language Independent Framework for Linguistic Code Switch Point Detection. The system uses characters level 5-grams and word level unigram language models to train a conditional random fields (CRF) model for classifying input words into various languages. We test our proposed framework and compare it to the state-of-the-art published systems on standard data sets from several language pairs: English-Spanish, Nepali-English, English-Hindi, Arabizi (Refers to Arabic written using the Latin/Roman script)-English, Arabic-Engari (Refers to English written using Arabic script), Modern Standard Arabic(MSA)-Egyptian, Levantine-MSA, Gulf-MSA, one more English-Spanish, and one more MSA-EGY. The overall weighted average F-score of each language pair are 96.4%, 97.3%, 98.0%, 97.0%, 98.9%, 86.3%, 88.2%, 90.6%, 95.2%, and 85.0% respectively. The results show that our approach despite its simplicity, either outperforms or performs at comparable levels to state-of-the-art published systems.
We introduce an electronic three-way lexicon, Tharwa, comprising Dialectal Arabic, Modern Standard Arabic and English correspondents. The paper focuses on Egyptian Arabic as the first pilot dialect for the resource, with plans to expand to other dialects of Arabic in later phases of the project. We describe Tharwas creation process and report on its current status. The lexical entries are augmented with various elements of linguistic information such as POS, gender, rationality, number, and root and pattern information. The lexicon is based on a compilation of information from both monolingual and bilingual existing resources such as paper dictionaries and electronic, corpus-based dictionaries. Multiple levels of quality checks are performed on the output of each step in the creation process. The importance of this lexicon lies in the fact that it is the first resource of its kind bridging multiple variants of Arabic with English. Furthermore, it is a wide coverage lexical resource containing over 73,000 Egyptian entries. Tharwa is publicly available. We believe it will have a significant impact on both Theoretical Linguistics as well as Computational Linguistics research.
In this paper, we present MADAMIRA, a system for morphological analysis and disambiguation of Arabic that combines some of the best aspects of two previously commonly used systems for Arabic processing, MADA (Habash and Rambow, 2005; Habash et al., 2009; Habash et al., 2013) and AMIRA (Diab et al., 2007). MADAMIRA improves upon the two systems with a more streamlined Java implementation that is more robust, portable, extensible, and is faster than its ancestors by more than an order of magnitude. We also discuss an online demo (see http://nlp.ldeo.columbia.edu/madamira/) that highlights these aspects.
The computational treatment of subjectivity and sentiment in natural language is usually significantly improved by applying features exploiting lexical resources where entries are tagged with semantic orientation (e.g., positive, negative values). In spite of the fair amount of work on Arabic sentiment analysis over the past few years (e.g., (Abbasi et al., 2008; Abdul-Mageed et al., 2014; Abdul-Mageed et al., 2012; Abdul-Mageed and Diab, 2012a; Abdul-Mageed and Diab, 2012b; Abdul-Mageed et al., 2011a; Abdul-Mageed and Diab, 2011)), the language remains under-resourced as to these polarity repositories compared to the English language. In this paper, we report efforts to build and present SANA, a large-scale, multi-genre, multi-dialect multi-lingual lexicon for the subjectivity and sentiment analysis of the Arabic language and dialects.
This tutorial introduces the different challenges and current solutions to the automatic processing of Arabic and its dialects. The tutorial has two parts: First, we present a discussion of generic issues relevant to Arabic NLP and detail dialectal linguistic issues and the challenges they pose for NLP. In the second part, we review the state-of-the-art in Arabic processing covering several enabling technologies and applications, e.g., dialect identification, morphological processing (analysis, disambiguation, tokenization, POS tagging), parsing, and machine translation.
Dialectal Arabic (DA) refers to the day-to-day vernaculars spoken in the Arab world. DA lives side-by-side with the official language, Modern Standard Arabic (MSA). DA differs from MSA on all levels of linguistic representation, from phonology and morphology to lexicon and syntax. Unlike MSA, DA has no standard orthography since there are no Arabic dialect academies, nor is there a large edited body of dialectal literature that follows the same spelling standard. In this paper, we present CODA, a conventional orthography for dialectal Arabic; it is designed primarily for the purpose of developing computational models of Arabic dialects. We explain the design principles of CODA and provide a detailed description of its guidelines as applied to Egyptian Arabic.
The Arabic language is a collection of dialectal variants along with the standard form, Modern Standard Arabic (MSA). MSA is used in official Settings while the dialectal variants (DA) correspond to the native tongue of the Arabic speakers. Arabic speakers typically code switch between DA and MSA, which is reflected extensively in written online social media. Automatic processing such Arabic genre is very difficult for automated NLP tools since the linguistic difference between MSA and DA is quite profound. However, no annotated resources exist for marking the regions of such switches in the utterance. In this paper, we present a simplified Set of guidelines for detecting code switching in Arabic on the word/token level. We use these guidelines in annotating a corpus that is rich in DA with frequent code switching to MSA. We present both a quantitative and qualitative analysis of the annotations.
Social relations like power and influence are difficult concepts to define, but are easily recognizable when expressed. In this paper, we describe a multi-layer annotation scheme for social power relations that are recognizable from online written interactions. We introduce a typology of four types of power relations between dialog participants: hierarchical power, situational power, influence and control of communication. We also present a corpus of Enron emails comprising of 122 threaded conversations, manually annotated with instances of these power relations between participants. Our annotations also capture attempts at exercise of power or influence and whether those attempts were successful or not. In addition, we also capture utterance level annotations for overt display of power. We describe the annotation definitions using two example email threads from our corpus illustrating each type of power relation. We also present detailed instructions given to the annotators and provide various statistics on annotations in the corpus.
We present AWATIF, a multi-genre corpus of Modern Standard Arabic (MSA) labeled for subjectivity and sentiment analysis (SSA) at the sentence level. The corpus is labeled using both regular as well as crowd sourcing methods under three different conditions with two types of annotation guidelines. We describe the sub-corpora constituting the corpus and provide examples from the various SSA categories. In the process, we present our linguistically-motivated and genre-nuanced annotation guidelines and provide evidence showing their impact on the labeling task.
In this paper, we present the details of creating a pilot Arabic proposition bank (Propbank). Propbanks exist for both English and Chinese. However the morphological and syntactic expression of linguistic phenomena in Arabic yields a very different type of process in creating an Arabic propbank. Hence, we highlight those characteristics of Arabic that make creating a propbank for the language a different challenge compared to the creation of an English Propbank.We believe that many of the lessons learned in dealing with Arabic could generalise to other languages that exhibit equally rich morphology and relatively free word order.
In this paper, we describe the methodological procedures and issues that emerged from the development of a pilot Levantine Arabic Treebank (LATB) at the Linguistic Data Consortium (LDC) and its use at the Johns Hopkins University (JHU) Center for Language and Speech Processing workshop on Parsing Arabic Dialects (PAD). This pilot, consisting of morphological and syntactic annotation of approximately 26,000 words of Levantine Arabic conversational telephone speech, was developed under severe time constraints; hence the LDC team drew on their experience in treebanking Modern Standard Arabic (MSA) text. The resulting Levantine dialect treebanked corpus was used by the PAD team to develop and evaluate parsers for Levantine dialect texts. The parsers were trained on MSA resources and adapted using dialect-MSA lexical resources (some developed especially for this task) and existing linguistic knowledge about syntactic differences between MSA and dialect. The use of the LATB for development and evaluation of syntactic parsers allowed the PAD team to provide feedbasck to the LDC treebank developers. In this paper, we describe the creation of resources for this corpus, as well as transformations on the corpus to eliminate speech effects and lessen the gap between our pre-existing MSA resources and the new dialectal corpus