Modeling Annotators: A Generative Approach to Learning from Annotator Rationales

By Omar Zaidan, Jason Eisner. 2008

175 citations recorded by the source corpus.

A human annotator can provide hints to a machine learner by highlighting contextual "rationales" for each of his or her annotations (Zaidan et al., 2007). How can one exploit this side information to better learn the desired parameters θ? We present a generative model of how a given annotator, knowing the true θ, stochastically chooses rationales. Thus, observing the rationales helps us infer the true θ. We collect substring rationales for a sentiment classification task (Pang and Lee, 2004) and use them to obtain significant accuracy improvements for each annotator. Our new generative approach exploits the rationales more effectively than our previous "masking SVM" approach. It is also more principled, and could be adapted to help learn other kinds of probabilistic classifiers for quite different tasks.

Subfields: Artificial Intelligence, Natural Language Processing, Signal Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Natural Language Processing

In Signal Processing

Citations in the corpus

Snorkel: Rapid Training Data Creation with Weak Supervision (2017, background): “Other forms include crowdsourced labels [37, 50], rules and heuristics for labeling data [39, 52], and others [29, 30, 30, 46, 51].”
ALICE: Active Learning with Contrastive Natural Language Explanations (2020, neutral): “As an estimate, Zhou et al. (2020); Hancock et al.”
Training Classifiers with Natural Language Explanations (2018, background): “text (Zaidan and Eisner, 2008; Arora and Nyberg, 2009), or marking relevant regions in images (Ahn et al.”
SELFEXPLAIN: A Self-Explaining Architecture for Neural Text Classifiers (2021, neutral): “, 2016) through expert annotations (Zaidan and Eisner, 2008).”
Measuring Association Between Labels and Free-Text Rationales (2020, background): “…SNLI (Bowman et al., 2015), SST (Socher et al., 2013), AgNews (Del Corso et al., 2005), Evidence Inference (Lehman et al., 2019), Movie Reviews (Zaidan and Eisner, 2008), MultiRC (Khashabi et al., 2018), LGD (Linzen et al., 2016), 20 News (Lang, 1995), Amazon Re-views (McAuley and Leskovec,…”
WT5?! Training Text-to-Text Models to Explain their Predictions (2020, background): “Movie Reviews (Zaidan & Eisner, 2008) is a sentiment analysis dataset where the goal is to predict whether a movie review has a positive or negative sentiment.”
Learning the Difference that Makes a Difference with Counterfactually-Augmented Data (2019, background): “Some ask annotators to highlight rationales, spans of text indicative of the label [28, 29, 24].”
Model-Agnostic Interpretability of Machine Learning (2016, background): “Feature labeling (Druck et al., 2008) or annotator rationales (Zaidan & Eisner, 2008) are other forms of feedback that should be supported for explanations.”
Regularizing Text Categorization with Clusters of Words (2016, methodology): “For the LDA, LSI, sentence, graph-of-words (GoW), word2vec regularizers, we perform grid search on the same set of values as ridge and lasso experiments for the ρ, λglas, λlas parameters.”
ERASER: A Benchmark to Evaluate Rationalized NLP Models (2019, background): “Work on learning from rationales that have been explicitly provided by users for text classification dates back over a decade (Zaidan et al., 2007; Zaidan and Eisner, 2008).”
Interpretable Neural Predictions with Differentiable Binary Variables (2019, methodology): “Our work is different from approaches that aim to improve clas-siﬁcation using rationales as an additional input (Zaidan et al., 2007; Zaidan and Eisner, 2008; Zhang et al., 2016).”
Towards Faithfully Interpretable NLP Systems: How Should We Define and Evaluate Faithfulness? (2020, background): “ample, consider the case of interpretation via posthoc text generation—where an additional “generator” component outputs a textual explanation of the model’s decision, and the generator is learned with supervision of textual explanations (Zaidan and Eisner, 2008; Rajani et al., 2019; Strout et al., ”
Multi-Level Structured Models for Document-Level Sentiment Classification (2010, methodology): “We use the movie reviews dataset from Zaidan et al. (2007) that was originally released by Pang and Lee (2004).”
Interactive Feature Space Construction using Semantic Information (2009, methodology): “These methods allow the annotator to directly specify information about the feature space in addition to providing labels, which is then incorporated into the learning algorithm (Huang and Mitchell, 2006; Raghavan and Allan, 2007; Zaidan et al., 2007; Druck et al., 2008; Zaidan and Eisner, 2008).”
Explanatory Interactive Machine Learning (2019, unknown): “Morerecently,researchersexploredfeaturesupervision[1,10, 11,23,24,30]andrationales[32,40,41],whichleveragebothlabel-andfeature-level(orsentence-level,forrationales)supervisionwith theaimofimprovinglearningefficiency.”
Explaining Patterns in Data with Language Models via Interpretable Autoprompting (2022, methodology): “…interpretation A popular method for interpreting neural networks is to inspect an LLM’s individual predictions via feature importances [29, 30], feature-interaction importances [31, 32], extractive rationales [33, 34], or natural-language explanations for individual predictions [35, 36].”