When we read a story, we bring to it a large body of implicit knowledge about the physical world. For instance, given the context “on stage, a woman takes a seat at the piano,” shown in Table 1, we can easily infer what the situation might look like: a woman is giving a piano performance, with a crowd watching her. We can furthermore infer her likely next action: she will most likely set her fingers on the piano keys and start playing.
This type of natural language inference requires commonsense reasoning, substantially broadening the scope of prior work that focused primarily on linguistic entailment Chierchia and McConnell-Ginet (2000). Whereas the dominant entailment paradigm asks if two natural language sentences (the ‘premise’ and the ‘hypothesis’) describe the same set of possible worlds Dagan et al. (2006); Bowman et al. (2015), here we focus on whether a (multiple-choice) ending describes a possible (future) world that can be anticipated from the situation described in the premise, even when it is not strictly entailed. Making such inference necessitates a rich understanding about everyday physical situations, including object affordances Gibson (1979) and frame semantics Baker et al. (1998).
On stage, a woman takes a seat at the piano. She
|a) sits on a bench as her sister plays with the doll.|
|b) smiles with someone as the music plays.|
|c) is in the crowd, watching the dancers.|
|d) nervously sets her fingers on the keys.|
A girl is going across a set of monkey bars. She
|a) jumps up across the monkey bars.|
|b) struggles onto the monkey bars to grab her head.|
|c) gets to the end and stands on a wooden plank.|
|d) jumps up and does a back flip.|
The woman is now blow drying the dog. The dog
|a) is placed in the kennel next to a woman’s feet.|
|b) washes her face with the shampoo.|
|c) walks into frame and walks towards the dog.|
|d) tried to cut her face, so she is trying to do something very close to her face.|
A first step toward grounded commonsense inference with today’s deep learning machinery is to create a large-scale dataset. However, recent work has shown that human-written datasets are susceptible toannotation artifacts: unintended stylistic patterns that give out clues for the gold labels Gururangan et al. (2018); Poliak et al. (2018)
. As a result, models trained on such datasets with human biases run the risk of over-estimating the actual performance on the underlying task, and are vulnerable to adversarial or out-of-domain examplesWang et al. (2018); Glockner et al. (2018).
In this paper, we introduce Adversarial Filtering (AF), a new method to automatically detect and reduce stylistic artifacts. We use this method to construct Swag: an adversarial dataset with 113k multiple-choice questions. We start with pairs of temporally adjacent video captions, each with a context and a follow-up event that we know is physically possible. We then use a state-of-the-art language model fine-tuned on this data to massively oversample a diverse set of possible negative sentence endings (or counterfactuals). Next, we filter these candidate endings aggressively and adversarially using a committee of trained models to obtain a population of de-biased endings with similar stylistic features to the real ones. Finally, these filtered counterfactuals are validated by crowd workers to further ensure data quality.
Extensive empirical results demonstrate unique contributions of our dataset, complementing existing datasets for natural langauge inference (NLI) Bowman et al. (2015); Williams et al. (2018) and commonsense reasoning Roemmele et al. (2011); Mostafazadeh et al. (2016); Zhang et al. (2017). First, our dataset poses a new challenge of grounded commonsense inference that is easy for humans (88%) while hard for current state-of-the-art NLI models (). Second, our proposed adversarial filtering methodology allows for cost-effective construction of a large-scale dataset while substantially reducing known annotation artifacts. The generality of adversarial filtering allows it to be applied to build future datasets, ensuring that they serve as reliable benchmarks.
2 Swag: Our new dataset
We introduce a new dataset for studying physically grounded commonsense inference, called Swag.111Short for Situations With Adversarial Generations. Our task is to predict which event is most likely to occur next in a video. More formally, a model is given a context : a complete sentence and a noun phrase that begins a second sentence, as well as a list of possible verb phrase sentence endings . See Figure 1 for an example triple . The model must then select the most appropriate verb phrase .
Our corpus consists of 113k multiple choice questions (73k training, 20k validation, 20k test) and is derived from pairs of consecutive video captions from ActivityNet Captions Krishna et al. (2017); Heilbron et al. (2015) and the Large Scale Movie Description Challenge (LSMDC; Rohrbach et al., 2017). The two datasets are slightly different in nature and allow us to achieve broader coverage: ActivityNet contains 20k YouTube clips containing one of 203 activity types (such as doing gymnastics or playing guitar); LSMDC consists of 128k movie captions (audio descriptions and scripts). For each pair of captions, we use a constituency parser Stern et al. (2017) to split the second sentence into noun and verb phrases (Figure 1).222We filter out sentences with rare tokens (3 occurrences), that are short (), or that lack a verb phrase. Each question has a human-verified gold ending and 3 distractors.
3 A solution to annotation artifacts
In this section, we outline the construction of Swag. We seek dataset diversity while minimizing annotation artifacts, conditional stylistic patterns such as length and word-preference biases. For many NLI datasets, these biases have been shown to allow shallow models (e.g. bag-of-words) obtain artificially high performance.
To avoid introducing easily “gamed” patterns, we present Adversarial Filtering (AF), a generally-applicable treatment involving the iterative refinement of a set of assignments to increase the entropy under a chosen model family. We then discuss how we generate counterfactual endings, and finally, the models used for filtering.
3.1 Formal definition
In this section, we formalize what it means for a dataset to be adversarial. Intuitively, we say that an adversarial dataset for a model is one on which will not generalize, even if evaluated on test data from the same distribution. More formally, let our input space be and the label space be . Our trainable classifier , taking parameters is defined as . Let our dataset of size be defined as
, and let the loss function over the dataset be. We say that a dataset is adversarial with respect to if we expect high empirical error over all leave-one-out train/test splits Vapnik (2000):
with regularization terms omitted for simplicity.
3.2 Adversarial filtering (AF) algorithm
In this section, we outline an approach for generating an adversarial dataset , effectively maximizing empirical error with respect to a family of trainable classifiers . Without loss of generality, we consider the situation where we have contexts, each associated with a single positive example , and a large population of context-specific negative examples , where for each . For instance, the negative examples could be incorrect relations in knowledge-base completion Socher et al. (2013), or all words in a dictionary for a single-word cloze task Zweig and Burges (2011).
Our goal will be to filter the population of negative examples for each instance to a size of . This will be captured by returning a set of assignments , where for each instance the assignment will be a -subset . The filtered dataset will then be:
Unfortunately, optimizing is difficult as is global and non-differentiable. To address this, we present Algorithm 1. On each iteration, we split the data into dummy ‘train’ and ‘test’ splits. We train a model on the training portion and obtain parameters , then use the remaining test portion to reassign the indices of . For each context, we replace some number of ‘easy’ negatives in that classifies correctly with ‘adversarial’ negatives outside of that misclassifies.
This process can be thought of as increasing the overall entropy of the dataset: given a strong model that is compatible with a random subset of the data, we aim to ensure it cannot generalize to the held-out set. We repeat this for several iterations to reduce the generalization ability of the model family over arbitrary train/test splits.
3.3 Generating candidate endings
To generate counterfactuals for Swag, we use an LSTM Hochreiter and Schmidhuber (1997) language model (LM), conditioned on contexts from video captions. We first pretrain on BookCorpus Zhu et al. (2015), then finetune on the video caption datasets. The architecture uses standard best practices and was validated on held-out perplexity of the video caption datasets; details are in the appendix. We use the LM to sample unique endings for a partial caption.333To ensure that the LM generates unique endings, we split the data into five validation folds and train five separate LMs, one for each set of training folds. This means that each LM never sees the found endings during training.
Importantly, we greedily sample the endings, since beam search decoding biases the generated endings to be of lower perplexity (and thus easily distinguishable from found endings). We find this process gives good counterfactuals: the generated endings tend to use topical words, but often make little sense physically, making them perfect for our task. Further, the generated endings are marked as “gibberish” by humans only 9.1% of the time (Sec 3.5); in that case the ending is filtered out.
3.4 Stylistic models for adversarial filtering
In creating Swag, we designed the model family to pick up on low-level stylistic features that we posit should not be predictive of whether an event happens next in a video. These stylistic features are an obvious case of annotation artifacts Cai et al. (2017); Schwartz et al. (2017).444A broad definition of annotation artifacts might include aspects besides lexical/stylistic features: for instance, certain events are less likely semantically regardless of the context (e.g. riding a horse using a hose). For this work, we erred more conservatively and only filtered based on style. Our final classifier is an ensemble of four stylistic models:
A multilayer perceptron (MLP) given LM perplexity features and context/ending lengths.
A bag-of-words model that averages the word embeddings of the second sentence as features.
A one-layer CNN, with filter sizes ranging from 2-5, over the second sentence.
A bidirectional LSTM over the 100 most common words in the second sentence; uncommon words are replaced by their POS tags.
We ensemble the models by concatenating their final representations and passing it through an MLP. On every adversarial iteration, the ensemble is trained jointly to minimize cross-entropy.
The accuracies of these models (at each iteration, evaluated on a 20% split of the test dataset before indices of get remapped) are shown in Figure 2. Performance decreases from 60% to close to random chance; moreover, confusing the perplexity-based MLP is not sufficient to lower performance of the ensemble. Only once the other stylistic models are added does the ensemble accuracy drop substantially, suggesting that our approach is effective at reducing stylistic artifacts.
3.5 Human verification
The final data-collection step is to have humans verify the data. Workers on Amazon Mechanical Turk were given the caption context, as well as six candidate endings: one found ending and five adversarially-sampled endings. The task was twofold: Turkers ranked the endings independently as likely, unlikely, or gibberish, and selected the best and second best endings (Fig 3).
We obtained the correct answers to each context in two ways. If a Turker ranks the found ending as either best or second best (73.7% of the time), we add the found ending as a gold example, with negatives from the generations not labelled best or gibberish. Further, if a Turker ranks a generated ending as best, and the found ending as second best, then we have reason to believe that the generation is good. This lets us add an additional training example, consisting of the generated best ending as the gold, and remaining generations as negatives.555These two examples share contexts. To prevent biasing the test and validation sets, we didn’t perform this procedure on answers from the evaluation sets’ context. Examples with non-gibberish endings were filtered out.666To be data-efficient, we reannotated filtered-out examples by replacing gibberish endings, as well as generations that outranked the found ending, with candidates from .
Label distribution by ending type
|Labels||Found end||Gen. end||ppa|
We found after 1000 examples that the annotators tended to have high agreement, also generally choosing found endings over generations (see Table 2). Thus, we collected the remaining 112k examples with one annotator each, periodically verifying that annotators preferred the found endings.
In this section, we evaluate the performance of various NLI models on Swag. Recall that models for our dataset take the following form: given a sentence and a noun phrase as context , as well as a list of possible verb phrase endings , a model must select a verb that hopefully matches :
To study the amount of bias in our dataset, we also consider models that take as input just the ending verb phrase , or the entire second sentence . For our learned models, we train
by minimizing multi-class cross-entropy. We consider three different types of word representations: 300d GloVe vectors from Common CrawlPennington et al. (2014), 300d Numberbatch vectors retrofitted using ConceptNet relations Speer et al. (2017), and 1024d ELMo contextual representations that show improvement on a variety of NLP tasks, including standard NLI Peters et al. (2018). We follow the final dataset split (see Section 2) using two training approaches: training on the found data, and the found and highly-ranked generated data. See the appendix for more details.
4.1 Unary models
The following models predict labels from a single span of text as input; this could be the ending only, the second sentence only, or the full passage.
fastText Joulin et al. (2017): This library models a single span of text as a bag of
-grams, and tries to predict the probability of an ending being correct or incorrect independently.777The fastText model is trained using binary cross-entropy; at test time we extract the prediction by selecting the ending with the highest positive likelihood under the model.
Pretrained sentence encoders We consider two types of pretrained RNN sentence encoders, SkipThoughts Kiros et al. (2015) and InferSent Conneau et al. (2017). SkipThoughts was trained by predicting adjacent sentences in book data, whereas InferSent was trained on supervised NLI data. For each second sentence (or just the ending), we feed the encoding into an MLP.
LSTM sentence encoder
Given an arbitrary span of text, we run a two-layer BiLSTM over it. The final hidden states are then max-pooled to obtain a fixed-size representation, which is then used to predict the potential for that ending.
|Ending only||2nd sentence only||Context+2nd sentence|
|found only||found+gen||found only||found+gen||found only||found+gen|
|LSTM sequence model||LSTM+GloVe||31.9||31.8||32.9||32.4||32.7||32.4||34.3||33.5||43.1||43.6||45.6||45.7|
|Dual sentence encoders||SkipThoughts-MLP||34.6||33.9||36.2||35.5||33.4||32.3||37.4||36.4|
|SNLI models (retrained)||DecompAttn+GloVe||29.8||30.3||31.1||31.7||47.4||47.6||48.5||48.6|
4.2 Binary models
The following models predict labels from two spans of text. We consider two possibilties for these models: using just the second sentence, where the two text spans are , or using the context and the second sentence, in which case the spans are . The latter case includes many models developed for the NLI task.
Dual Bag-of-Words For this baseline, we treat each sentence as a bag-of-embeddings . We model the probability of picking an ending using a bilinear model: .888We also tried using an MLP, but got worse results.
Dual pretrained sentence encoders Here, we obtain representations from SkipThoughts or InferSent for each span, and compute their pairwise compatibility using either 1) a bilinear model or 2) an MLP from their concatenated representations.
SNLI inference Here, we consider two models that do well on SNLI Bowman et al. (2015): Decomposable Attention Parikh et al. (2016) and ESIM Chen et al. (2017). We use pretrained versions of these models (with ELMo embeddings) on SNLI to obtain 3-way entailment, neutral, and contradiction probabilities for each example. We then train a log-linear model using these 3-way probabilities as features.
SNLI models (retrained) Here, we train ESIM and Decomposable Attention on our dataset: we simply change the output layer size to 1 (the potential of an ending ) with a softmax over .
4.3 Other models
We also considered the following models:
Length: Although length was used by the adversarial classifier, we want to verify that human validation didn’t reintroduce a length bias. For this baseline, we always choose the shortest ending.
As our task requires world knowledge, we tried a rule-based system on top of the ConceptNet knowledge baseSpeer et al. (2017). For an ending sentence, we use the spaCy dependency parser to extract the head verb and its dependent object. The ending score is given by the number of ConceptNet causal relations999We used the relations ‘Causes’, ‘CapableOf’, ‘ReceivesAction’, ‘UsedFor’, and ‘HasSubevent’. Though their coverage is low (30.4% of questions have an answer with 1 causal relation), the more frequent relations in ConceptNet, such as ‘IsA’, at best only indirectly relate to our task. between synonyms of the verb and synonyms of the object.
Human performance To benchmark human performance, five Mechanical Turk workers were asked to answer 100 dataset questions, as did an ‘expert’ annotator (the first author of this paper). Predictions were combined using a majority vote.
We present our results in Table 3. The best model that only uses the ending is the LSTM sequence model with ELMo embeddings, which obtains 43.6%. This model, as with most models studied, greatly improves with more context: by 3.1% when given the initial noun phrase, and by an additional 4% when also given the first sentence.
Further improvement is gained from models that compute pairwise representations of the inputs. While the simplest such model, DualBoW, obtains only 35.1% accuracy, combining InferSent sentence representations gives 40.5% accuracy (InferSent-Bilinear). The best results come from pairwise NLI models: when fully trained on Swag, ESIM+ELMo obtains 59.2% accuracy.
When comparing machine results to human results, we see there exists a lot of headroom. Though there likely is some noise in the task, our results suggest that humans (even untrained) converge to a consensus. Our in-house “expert” annotator is outperformed by an ensemble of 5 Turk workers (with 88% accuracy); thus, the effective upper bound on our dataset is likely even higher.
5.1 Swag versus existing NLI datasets
The past few years have yielded great advances in NLI and representation learning, due to the availability of large datasets like SNLI and MultiNLI Bowman et al. (2015); Williams et al. (2018). With the release of Swag, we hope to continue this trend, particularly as our dataset largely has the same input/output format as other NLI datasets. We observe three key differences between our dataset and others in this space:
First, as noted in Section 1, Swag requires a unique type of temporal reasoning. A state-of-the-art NLI model such as ESIM, when bottlenecked through the SNLI notion of entailment (SNLI-ESIM), only obtains 36.1% accuracy.101010The weights of SNLI-ESIM pick up primarily on entailment probability (0.59), as with neutral (0.46), while contradiction is negatively correlated (-.42). This implies that these datasets necessitate different (and complementary) forms of reasoning.
Second, our use of videos results in wide coverage of dynamic and temporal situations Compared with SNLI, with contexts from Flickr30K Plummer et al. (2017) image captions, Swag has more active verbs like ‘pull’ and ‘hit,’ and fewer static verbs like ‘sit’ and ‘wear’ (Figure 4).111111Video data has other language differences; notably, character names in LSMDC were replaced by ‘someone’
Third, our dataset suffers from few lexical biases. Whereas fastText, a bag of -gram model, obtains 67.0% accuracy on SNLI versus a 34.3% baseline Gururangan et al. (2018), fastText obtains only 29.0% accuracy on Swag.121212The most predictive individual words on SWAG are infrequent in number: ‘dotted‘ with P with 10.3 counts, and P with 16.3 counts. (Counts from negative endings were discounted 3x, as there are 3 times as many negative endings as positive endings).
|Situational||The good ending is better in context.||53.7%|
|Plausibility||The bad ending is implausible regardless of context.||14.4%|
|Novelty||The bad ending seems redundant; it is entailed by the context.||1.8%|
|Weirdness||The bad ending is semantically or grammatically malformed, e.g. ‘the man is getting out of the horse.’||18.1%|
|Ambiguous||Both endings seem equally likely.||12.0%|
5.2 Error analysis
We sought to quantify how human judgments differ from the best studied model, ESIM+ELMo. We randomly sampled 100 validation questions that ESIM+ELMo answered incorrectly, for each extracting both the gold ending and the model’s preferred ending. We asked 5 Amazon Mechanical Turk workers to pick the better ending (of which they preferred the gold endings 94% of the time) and to select one (or more) multiple choice reasons explaining why the chosen answer was better.
The options, and the frequencies, are outlined in Table 4. The most common reason for the turkers preferring the correct answer is situational (52.3% of the time), followed by weirdness (17.5%) and plausibility (14.4%). This suggests that ESIM+ELMo already does a good job at filtering out weird and implausible answers, with the main bottleneck being grounded physical understanding. The ambiguous percentage is also relatively low (12.0%), implying significant headroom.
5.3 Qualitative examples
Last, we show several qualitative examples in Table 5. Though models can do decently well by identifying complex alignment patterns between the two sentences (e.g. being “up a tree” implies that “tree” is the end phrase), the incorrect model predictions suggest this strategy is insufficient. For instance, answering “An old man rides a small bumper car” requires knowledge about bumper cars and how they differ from regular cars: bumper cars are tiny, don’t drive on roads, and don’t work in parking lots, eliminating the alternatives. However, this knowledge is difficult to extract from existing corpora: for instance, the ConceptNet entry for Bumper Car has only a single relation: bumper cars are a type of vehicle. Other questions require intuitive physical reasoning: e.g, for “he pours the raw egg batter into the pan,” about what happens next in making an omelet.
5.4 Where to go next?
Our results suggest that Swag is a challenging testbed for NLI models. However, the adversarial models used to filter the dataset are purely stylistic and focus on the second sentence; thus, subtle artifacts still likely remain in our dataset. These patterns are ostensibly picked up by the NLI models (particularly when using ELMo features), but the large gap between machine and human performance suggests that more is required to solve the dataset. As models are developed for commonsense inference, and more broadly as the field of NLP advances, we note that AF can be used again to create a more adversarial version of Swag using better language models and AF models.
6 Related Work
There has been a long history of NLI benchmarks focusing on linguistic entailment Cooper et al. (1996); Dagan et al. (2006); Marelli et al. (2014); Bowman et al. (2015); Lai et al. (2017); Williams et al. (2018). Recent NLI datasets in particular have supported learning broadly-applicable sentence representations Conneau et al. (2017); moreover, models trained on these datasets were used as components for performing better video captioning Pasunuru and Bansal (2017), summarization Pasunuru and Bansal (2018), and generation Holtzman et al. (2018), confirming the importance of NLI research. The NLI task requires a variety of commonsense knowledge LoBue and Yates (2011), which our work complements. However, previous datasets for NLI have been challenged by unwanted annotation artifacts, Gururangan et al. (2018); Poliak et al. (2018) or scale issues. Our work addresses these challenges by constructing a new NLI benchmark focused on grounded commonsense reasoning, and by introducing an adversarial filtering mechanism that substantially reduces known and easily detectable annotation artifacts.
Several datasets have been introduced to study NLI beyond linguistic entailment: for inferring likely causes and endings given a sentence (COPA; Roemmele et al., 2011), for choosing the most sensible ending to a short story (RocStories; Mostafazadeh et al., 2016; Sharma et al., 2018), and for predicting likelihood of a hypothesis by regressing to an ordinal label (JOCI; Zhang et al. (2017)). These datasets are relatively small: 1k examples for COPA and 10k cloze examples for RocStories.131313For RocStories, this was by design to encourage learning from the larger corpus of 98k sensible stories.
JOCI increases the scale by generating the hypotheses using a knowledge graph or a neural model. In contrast to JOCI where the task was formulated as a regression task on the degree of plausibility of the hypothesis, we frame commonsense inference as a multiple choice question to reduce the potential ambiguity in the labels and to allow for direct comparison between machines and humans. In addition,Swag’s use of adversarial filtering increases diversity of situations and counterfactual generation quality.
Last, another related task formulation is sentence completion or cloze, where the task is to predict a single word that is removed from a given context Zweig and Burges (2011); Paperno et al. (2016).141414 Prior work on sentence completion filtered negatives with heuristics based on LM perplexities. We initially tried something similar, but found the result to still be gameable.
Prior work on sentence completion filtered negatives with heuristics based on LM perplexities. We initially tried something similar, but found the result to still be gameable.Our work in contrast requires longer textual descriptions to reason about.
Several resources have been introduced to study temporal inference in vision. The Visual Madlibs dataset has 20k image captions about hypothetical next/previous events Yu et al. (2015); similar to our work, the test portion is multiple-choice, with counterfactual answers retrieved from similar images and verified by humans. The question of ‘what will happen next?’ has also been studied in photo albums Huang et al. (2016), videos of team sports, Felsen et al. (2017) and egocentric dog videos Ehsani et al. (2018). Last, annotation artifacts are also a recurring problem for vision datasets such as Visual Genome Zellers et al. (2018) and Visual QA Jabri et al. (2016); recent work was done to create a more challenging VQA dataset by annotating complementary image pairs Goyal et al. (2016).
Reducing gender/racial bias
Prior work has sought to reduce demographic biases in word embeddings Zhang et al. (2018) as well as in image recognition models Zhao et al. (2017). Our work has focused on producing a dataset with minimal annotation artifacts, which in turn helps to avoid some gender and racial biases that stem from elicitation Rudinger et al. (2017). However, it is not perfect in this regard, particularly due to biases in movies Schofield and Mehr (2016); Sap et al. (2017). Our methodology could potentially be extended to construct datasets free of (possibly intersectional) gender or racial bias.
Prior work has studied learning grounded knowledge about objects and verbs: from knowledge bases Li et al. (2016), syntax parses Forbes and Choi (2017), word embeddings Lucy and Gauthier (2017), and images and dictionary definitions Zellers and Choi (2017). An alternate thread of work has been to learn scripts: high-level representations of event chains Schank and Abelson (1975); Chambers and Jurafsky (2009). Swag evaluates both of these strands.
We propose a new challenge of physically situated commonsense inference that broadens the scope of natural language inference (NLI) with commonsense reasoning. To support research toward commonsense NLI, we create a large-scale dataset Swag with 113k multiple-choice questions. Our dataset is constructed using Adversarial Filtering (AF), a new paradigm for robust and cost-effective dataset construction that allows datasets to be constructed at scale while automatically reducing annotation artifacts that can be easily detected by a committee of strong baseline models. Our adversarial filtering paradigm is general, allowing potential applications to other datasets that require human composition of question answer pairs.
We thank the anonymous reviewers, members of the ARK and xlab at the University of Washington, researchers at the Allen Institute for AI, and Luke Zettlemoyer for their helpful feedback. We also thank the Mechanical Turk workers for doing a fantastic job with the human validation. This work was supported by the National Science Foundation Graduate Research Fellowship (DGE-1256082), the NSF grant (IIS-1524371, 1703166), the DARPA CwC program through ARO (W911NF-15-1-0543), the IARPA DIVA program through D17PC00343, and gifts by Google and Facebook. The views and conclusions contained herein are those of the authors and should not be interpreted as representing endorsements of IARPA, DOI/IBC, or the U.S. Government.
Appendix A Appendix
a.1 More detail about video datasets
As mentioned in the main paper, we obtained contexts and found endings from video data. The videos in the ActivityNet dataset are already broken up into into clips. However, the LSMDC dataset contains captions for the entire movie, so it is possible that temporally adjacent captions describe events that are far apart in time. Thus, we don’t include any pair of captions that have a time-difference of more than 25 seconds.
In addition to the datasets we used, we also considered the DiDeMo dataset, which consists of (often several) referring expressions in a video Hendricks et al. (2017). However, many of the referring expressions are themselves sentence fragments, (e.g. “first time we see people” so we ultimately did not use this dataset.) Additionally, we considered the Visual Madlibs dataset Yu et al. (2015), as it contains 10k hypothetical captions written by Mechanical Turk workers about what might happen next given an image. However, these captions are fundamentally different from the rest of the data (as they’re about what might) happen next; as a result, they use different types of language. They also have different tenses versus the other datasets that we considered (e.g. past tense), as a result of the “Mad-libs” style of data collection.
a.2 Details of the language model
Our language model follows standard best practices: the input and output embedding layers are tied (Inan et al., 2017; Press and Wolf, 2017), all embedding and hidden layers are set to 512, and we used recurrent dropout Gal and Ghahramani (2016) on the hidden states and embedding layer. We additionally train a backwards
language model alongside the forward language model, and they share embedding parameters. This adds extra supervision to the embedding layer and gives us another way to score candidate generations. We first pretrain the language model for two epochs on pairs of two sentences in the Toronto Books datasetZhu et al. (2015), and then train on sentence pairs from ActivityNet Captions and LSMDC, validating on held-out perplexity. For optimization, we use Adam Kingma and Ba (2015) with a learning rate of and clip gradients to norm .
All of the above details were validated using perplexity on a held-out set of the video datasets during early experimentation. Our final development set forward perplexity was 31.2 and backward perplexity was 30.4. We tried more complicated language modeling architectures, such as from Józefowicz et al. (2016), but ended up not seeing an improvement due to overfitting.
a.3 Language model features for the MLP, during adversarial filtering
We obtained LM perplexity features to be used during adversarial filtering in the following ways, using both directions of the bidirectional language model. We extract perplexities for the context by itself (going forward), the ending given the context (going forward), the context given the ending (going backward), and the ending by itself (going backward). We also extract the probability of the final generated token going forward, since sentences sometimes reach the length limit of 25 tokens and end unnaturally.
a.4 Refinining the generated answers to four distractors
In the main paper, we noted that we started with 1023 negatives per example, which the adversarial filtering process filtered down to 9. Five of these were passed to mechanical turk workers, and we were left with anywhere between 0 and 4 of these per example as “distractors.” (Note that we always were filtering out the second best option that the was selected by the turkers). This means that for many of our examples (62%) we actually have a fourth distractor. In these cases, we sorted the distractors by their “unlikely/likely” score, so that the fourth distractor was the one deemed most likely. We still provided the fourth distractor in the training set to be possibly used in future work, however we didn’t train on it for simplicity.
a.5 More information about Mechanical turk
We used several tricks to keep the interannotator agreement high (with a pairwise percent agreement of 79% at classifying an ending as either in the Top 2). First, we had a screening HIT where turkers were given detailed instructions for the task, and only the best-scoring turk workers qualified for the remaining HITs. Second, we periodically dequalified turkers that had a low agreement with the gold endings: any turk worker with an accuracy of less than 55% of classifying the “gold” ending as the best or second best, over 10 or more HITs, had the qualification taken away. We also gave small bonuses to turkers with high accuracy.
During our crowdsourcing, we tried to pay the Turkers a fair wage (median $8.57 per hour) and they left positive comments for us on TurkOpticon and TurkerView. The total dataset cost was $23,000, or an average of 20 cents per example.
a.6 Implementation details of the models considered
We implemented the neural models in PyTorch using the AllenNLP libraryGardner et al. (2018). Our experiments use the Adam optimizer Kingma and Ba (2015), with a learning rate of
and gradient clipping, except for Decomposable Attention and ESIM, where we use the AllenNLP default configurations.
|Questions with only generated endings||25,618|
|Questions with one original ending||87,939|
|Questions in total||113,557|
|Sentence pairs from ActivityNet||51,439|
|Sentence pairs from LSMDC||62,118|
a.7 More info about dataset diversity
The final dataset has a vocabulary size of 21000. We also visualize the coverage of the dataset with a Topic model (see Table 7).
|5.0%||ball, pull, hit, wall, inside, time, game, rope, team|
|4.9%||window, red, long, drink, bowl, ingredient, mix|
|6.1%||arm, speak, appear, climb, tree, roll, like, roof, edge|
|4.0%||water, bar, board, blue, boat, fly, river, join, dive|
|5.3%||eye, smile, close, little, lean, cover, remove, lip|
|4.6%||walk, outside, street, wave, pass, beach, sidewalk|
|5.7%||field, drop, slide, drive, right, kick, park, road, chest|
|4.7%||watch, dog, flip, stick, land, demonstrate, trick, mat|
|4.5%||dance, lift, try, line, snow, gun, catch, hill, bend|
|4.6%||fall, crowd, pour, shake, finish, raise, grass, wooden|
|5.9%||perform, spin, house, stage, routine, fence, bow|
a.8 Comparing the distribution of verbs with MultiNLI
We also produced an extension to Figure 4 of the main paper, that involves verbs from MultiNLI, in Figure 5. We ended up not including it in the paper because we wanted to focus our comparison between SNLI and Swag (as they are both grounded datasets). Interestingly, we find that Swag
has a less skewed cumulative distribution of verbs up to around 120, when afterwards MultiNLI has a slightly less skewed distribution. This is possibly due to the broader set of domains considered by MultiNLI, whereas we consider videos (which is also a broad domain! but still underrepresents words highly used in newswire text, for instance.)
a.9 More examples
We have more qualitative examples in Table 8.
- Baker et al. (1998) Collin F Baker, Charles J Fillmore, and John B Lowe. 1998. The berkeley framenet project. In Proceedings of the 17th international conference on Computational linguistics-Volume 1, pages 86–90. Association for Computational Linguistics.
Blei et al. (2003)
David M Blei, Andrew Y Ng, and Michael I Jordan. 2003.
Latent dirichlet allocation.
Journal of machine Learning research, 3(Jan):993–1022.
Bowman et al. (2015)
Samuel R. Bowman, Gabor Angeli, Christopher Potts, and Christopher D. Manning.
A large annotated corpus for learning natural language inference.
Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, EMNLP 2015, Lisbon, Portugal, September 17-21, 2015, pages 632–642.
- Cai et al. (2017) Zheng Cai, Lifu Tu, and Kevin Gimpel. 2017. Pay attention to the ending: Strong neural baselines for the roc story cloze task. In Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), volume 2, pages 616–622.
- Chambers and Jurafsky (2009) Nathanael Chambers and Dan Jurafsky. 2009. Unsupervised Learning of Narrative Schemas and Their Participants. In Proceedings of the Joint Conference of the 47th Annual Meeting of the ACL and the 4th International Joint Conference on Natural Language Processing of the AFNLP: Volume 2 - Volume 2, ACL ’09, pages 602–610, Stroudsburg, PA, USA. Association for Computational Linguistics.
- Chen et al. (2017) Qian Chen, Xiaodan Zhu, Zhen-Hua Ling, Si Wei, Hui Jiang, and Diana Inkpen. 2017. Enhanced lstm for natural language inference. In Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), volume 1, pages 1657–1668.
- Chierchia and McConnell-Ginet (2000) Gennaro Chierchia and Sally McConnell-Ginet. 2000. Meaning and Grammar (2Nd Ed.): An Introduction to Semantics. MIT Press, Cambridge, MA, USA.
- Conneau et al. (2017) Alexis Conneau, Douwe Kiela, Holger Schwenk, Loïc Barrault, and Antoine Bordes. 2017. Supervised learning of universal sentence representations from natural language inference data. In Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, pages 670–680.
- Cooper et al. (1996) Robin Cooper, Dick Crouch, JV Eijckl, Chris Fox, JV Genabith, J Japars, Hans Kamp, David Milward, Manfred Pinkal, Massimo Poesio, et al. 1996. A framework for computational semantics (fracas). Technical report, Technical report, The FraCaS Consortium.
- Dagan et al. (2006) Ido Dagan, Oren Glickman, and Bernardo Magnini. 2006. The pascal recognising textual entailment challenge. In Machine learning challenges. evaluating predictive uncertainty, visual object classification, and recognising tectual entailment, pages 177–190. Springer.
- Ehsani et al. (2018) Kiana Ehsani, Hessam Bagherinezhad, Joseph Redmon, Roozbeh Mottaghi, and Ali Farhadi. 2018. Who let the dogs out? modeling dog behavior from visual data. In
- Felsen et al. (2017) Panna Felsen, Pulkit Agrawal, and Jitendra Malik. 2017. What will happen next? forecasting player moves in sports videos. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 3342–3351.
- Forbes and Choi (2017) Maxwell Forbes and Yejin Choi. 2017. Verb physics: Relative physical knowledge of actions and objects. In Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), volume 1, pages 266–276.
Gal and Ghahramani (2016)
Yarin Gal and Zoubin Ghahramani. 2016.
A theoretically grounded application of dropout in recurrent neural networks.In Advances in neural information processing systems, pages 1019–1027.
- Gardner et al. (2018) Matt Gardner, Joel Grus, Mark Neumann, Oyvind Tafjord, Pradeep Dasigi, Nelson Liu, Matthew Peters, Michael Schmitz, and Luke Zettlemoyer. 2018. Allennlp: A deep semantic natural language processing platform. arXiv preprint arXiv:1803.07640.
- Gibson (1979) JJ Gibson. 1979. The ecological approach to visual perception. Houghton Mifflin Comp.
- Glockner et al. (2018) Max Glockner, Vered Shwartz, and Yoav Goldberg. 2018. Breaking nli systems with sentences that require simple lexical inferences. In Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), pages 650–655, Melbourne, Australia. Association for Computational Linguistics.
- Goyal et al. (2016) Yash Goyal, Tejas Khot, Douglas Summers-Stay, Dhruv Batra, and Devi Parikh. 2016. Making the V in VQA matter: Elevating the role of image understanding in Visual Question Answering. arXiv preprint arXiv:1612.00837.
- Gururangan et al. (2018) Suchin Gururangan, Swabha Swayamdipta, Omer Levy, Roy Schwartz, Samuel Bowman, and Noah A. Smith. 2018. Annotation artifacts in natural language inference data. In Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 2 (Short Papers), pages 107–112. Association for Computational Linguistics.
- Heilbron et al. (2015) Fabian Caba Heilbron, Victor Escorcia, Bernard Ghanem, and Juan Carlos Niebles. 2015. Activitynet: A large-scale video benchmark for human activity understanding. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 961–970.
Hendricks et al. (2017)
Lisa Anne Hendricks, Oliver Wang, Eli Shechtman, Josef Sivic, Trevor Darrell,
and Bryan Russell. 2017.
Localizing moments in video with natural language.In Proceedings of the IEEE International Conference on Computer Vision (ICCV).
- Hochreiter and Schmidhuber (1997) Sepp Hochreiter and Jürgen Schmidhuber. 1997. Long short-term memory. Neural Comput., 9(8):1735–1780.
- Holtzman et al. (2018) Ari Holtzman, Jan Buys, Maxwell Forbes, Antoine Bosselut, David Golub, and Yejin Choi. 2018. Learning to write with cooperative discriminators. In Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 1638–1649. Association for Computational Linguistics.
- Huang et al. (2016) Ting-Hao Kenneth Huang, Francis Ferraro, Nasrin Mostafazadeh, Ishan Misra, Aishwarya Agrawal, Jacob Devlin, Ross Girshick, Xiaodong He, Pushmeet Kohli, Dhruv Batra, et al. 2016. Visual storytelling. In Proceedings of the 2016 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 1233–1239.
- Inan et al. (2017) Hakan Inan, Khashayar Khosravi, and Richard Socher. 2017. Tying word vectors and word classifiers: A loss framework for language modeling. In ICLR. ICLR.
- Jabri et al. (2016) Allan Jabri, Armand Joulin, and Laurens van der Maaten. 2016. Revisiting visual question answering baselines. In European conference on computer vision, pages 727–739. Springer.
- Joulin et al. (2017) Armand Joulin, Edouard Grave, Piotr Bojanowski, and Tomas Mikolov. 2017. Bag of tricks for efficient text classification. In Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguistics: Volume 2, Short Papers, volume 2, pages 427–431.
- Józefowicz et al. (2016) Rafal Józefowicz, Oriol Vinyals, Mike Schuster, Noam Shazeer, and Yonghui Wu. 2016. Exploring the limits of language modeling. CoRR, abs/1602.02410.
- Kingma and Ba (2015) Diederik P. Kingma and Jimmy Ba. 2015. Adam: A method for stochastic optimization. In ICLR. ICLR.
- Kiros et al. (2015) Ryan Kiros, Yukun Zhu, Ruslan R Salakhutdinov, Richard Zemel, Raquel Urtasun, Antonio Torralba, and Sanja Fidler. 2015. Skip-thought vectors. In Advances in neural information processing systems, pages 3294–3302.
- Krishna et al. (2017) Ranjay Krishna, Kenji Hata, Frederic Ren, Li Fei-Fei, and Juan Carlos Niebles. 2017. Dense-Captioning Events in Videos. In International Conference on Computer Vision (ICCV).
- Lai et al. (2017) Alice Lai, Yonatan Bisk, and Julia Hockenmaier. 2017. Natural language inference from multiple premises. In Proceedings of the Eighth International Joint Conference on Natural Language Processing (Volume 1: Long Papers), pages 100–109, Taipei, Taiwan. Asian Federation of Natural Language Processing.
- Li et al. (2016) Xiang Li, Aynaz Taheri, Lifu Tu, and Kevin Gimpel. 2016. Commonsense knowledge base completion. In Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 1445–1455, Berlin, Germany. Association for Computational Linguistics.
- LoBue and Yates (2011) Peter LoBue and Alexander Yates. 2011. Types of common-sense knowledge needed for recognizing textual entailment. In Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies: short papers-Volume 2, pages 329–334. Association for Computational Linguistics.
Lucy and Gauthier (2017)
Li Lucy and Jon Gauthier. 2017.
Are distributional representations ready for the real world? evaluating word vectors for grounded perceptual meaning.In Proceedings of the First Workshop on Language Grounding for Robotics, pages 76–85.
- Marelli et al. (2014) Marco Marelli, Stefano Menini, Marco Baroni, Luisa Bentivogli, Raffaella bernardi, and Roberto Zamparelli. 2014. A sick cure for the evaluation of compositional distributional semantic models. In Proceedings of the Ninth International Conference on Language Resources and Evaluation (LREC’14), pages 216–223, Reykjavik, Iceland. European Language Resources Association (ELRA). ACL Anthology Identifier: L14-1314.
- Mostafazadeh et al. (2016) Nasrin Mostafazadeh, Nathanael Chambers, Xiadong He, Devi Parikh, Dhruv Batra, Lucy Vanderwende, Pushmeet Kohli, and James Allen. 2016. A corpus and evaluation framework for deeper understanding of commonsense stories. In NAACL.
- Paperno et al. (2016) Denis Paperno, Germán Kruszewski, Angeliki Lazaridou, Ngoc Quan Pham, Raffaella Bernardi, Sandro Pezzelle, Marco Baroni, Gemma Boleda, and Raquel Fernandez. 2016. The lambada dataset: Word prediction requiring a broad discourse context. In Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 1525–1534, Berlin, Germany. Association for Computational Linguistics.
Parikh et al. (2016)
Ankur Parikh, Oscar Täckström, Dipanjan Das, and Jakob Uszkoreit. 2016.
A decomposable attention model for natural language inference.In Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing, pages 2249–2255.
- Pasunuru and Bansal (2017) Ramakanth Pasunuru and Mohit Bansal. 2017. Multi-task video captioning with video and entailment generation. In Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 1273–1283, Vancouver, Canada. Association for Computational Linguistics.
- Pasunuru and Bansal (2018) Ramakanth Pasunuru and Mohit Bansal. 2018. Multi-reward reinforced summarization with saliency and entailment. In Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 2 (Short Papers), pages 646–653. Association for Computational Linguistics.
- Pennington et al. (2014) Jeffrey Pennington, Richard Socher, and Christopher Manning. 2014. Glove: Global vectors for word representation. In Proceedings of the 2014 conference on empirical methods in natural language processing (EMNLP), pages 1532–1543.
- Peters et al. (2018) Matthew Peters, Mark Neumann, Mohit Iyyer, Matt Gardner, Christopher Clark, Kenton Lee, and Luke Zettlemoyer. 2018. Deep contextualized word representations. In Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Papers), pages 2227–2237. Association for Computational Linguistics.
- Plummer et al. (2017) Bryan A. Plummer, Liwei Wang, Chris M. Cervantes, Juan C. Caicedo, Julia Hockenmaier, and Svetlana Lazebnik. 2017. Flickr30k entities: Collecting region-to-phrase correspondences for richer image-to-sentence models. Int. J. Comput. Vision, 123(1):74–93.
- Poliak et al. (2018) Adam Poliak, Jason Naradowsky, Aparajita Haldar, Rachel Rudinger, and Benjamin Van Durme. 2018. Hypothesis Only Baselines in Natural Language Inference. In Joint Conference on Lexical and Computational Semantics (StarSem).
- Press and Wolf (2017) Ofir Press and Lior Wolf. 2017. Using the output embedding to improve language models. In Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguistics: Volume 2, Short Papers, volume 2, pages 157–163.
- Roemmele et al. (2011) Melissa Roemmele, Cosmin Adrian Bejan, and Andrew S Gordon. 2011. Choice of Plausible Alternatives: An Evaluation of Commonsense Causal Reasoning. In AAAI Spring Symposium: Logical Formalizations of Commonsense Reasoning.
- Rohrbach et al. (2017) Anna Rohrbach, Atousa Torabi, Marcus Rohrbach, Niket Tandon, Christopher Pal, Hugo Larochelle, Aaron Courville, and Bernt Schiele. 2017. Movie Description. International Journal of Computer Vision, 123(1):94–120.
- Rudinger et al. (2017) Rachel Rudinger, Chandler May, and Benjamin Van Durme. 2017. Social bias in elicited natural language inferences. In Proceedings of the First ACL Workshop on Ethics in Natural Language Processing, pages 74–79.
- Sap et al. (2017) Maarten Sap, Marcella Cindy Prasettio, Ari Holtzman, Hannah Rashkin, and Yejin Choi. 2017. Connotation frames of power and agency in modern films. In Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, pages 2329–2334.
- Schank and Abelson (1975) Roger C. Schank and Robert P. Abelson. 1975. Scripts, plans, and knowledge. In Proceedings of the 4th International Joint Conference on Artificial Intelligence - Volume 1, IJCAI’75, pages 151–157, San Francisco, CA, USA. Morgan Kaufmann Publishers Inc.
- Schofield and Mehr (2016) Alexandra Schofield and Leo Mehr. 2016. Gender-distinguishing features in film dialogue. In Proceedings of the Fifth Workshop on Computational Linguistics for Literature, pages 32–39.
- Schwartz et al. (2017) Roy Schwartz, Maarten Sap, Ioannis Konstas, Li Zilles, Yejin Choi, and Noah A. Smith. 2017. The effect of different writing tasks on linguistic style: A case study of the ROC story cloze task. In Proc. of CoNLL.
- Sharma et al. (2018) Rishi Sharma, James Allen, Omid Bakhshandeh, and Nasrin Mostafazadeh. 2018. Tackling the story ending biases in the story cloze test. In Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), volume 2, pages 752–757.
Socher et al. (2013)
Richard Socher, Danqi Chen, Christopher D Manning, and Andrew Ng. 2013.
Reasoning with neural tensor networks for knowledge base completion.In Advances in neural information processing systems, pages 926–934.
- Speer et al. (2017) Robert Speer, Joshua Chin, and Catherine Havasi. 2017. Conceptnet 5.5: An open multilingual graph of general knowledge. In AAAI Conference on Artificial Intelligence, pages 4444–4451.
- Stern et al. (2017) Mitchell Stern, Jacob Andreas, and Dan Klein. 2017. A minimal span-based neural constituency parser. In Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), volume 1, pages 818–827.
Vladimir Vapnik. 2000.
The Nature of Statistical Learning Theory, 2 edition. Information Science and Statistics. Springer-Verlag, New York.
- Wang et al. (2018) Alex Wang, Amapreet Singh, Julian Michael, Felix Hill, Omer Levy, and Samuel R Bowman. 2018. Glue: A multi-task benchmark and analysis platform for natural language understanding. arXiv preprint arXiv:1804.07461.
- Williams et al. (2018) Adina Williams, Nikita Nangia, and Samuel Bowman. 2018. A broad-coverage challenge corpus for sentence understanding through inference. In Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Papers), pages 1112–1122. Association for Computational Linguistics.
- Yu et al. (2015) Licheng Yu, Eunbyung Park, Alexander C. Berg, and Tamara L. Berg. 2015. Visual Madlibs: Fill in the blank Image Generation and Question Answering. ICCV.
- Zellers and Choi (2017) Rowan Zellers and Yejin Choi. 2017. Zero-shot activity recognition with verb attribute induction. In Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing (EMNLP).
- Zellers et al. (2018) Rowan Zellers, Mark Yatskar, Sam Thomson, and Yejin Choi. 2018. Neural motifs: Scene graph parsing with global context. In Conference on Computer Vision and Pattern Recognition.
- Zhang et al. (2018) Brian Hu Zhang, Blake Lemoine, and Margaret Mitchell. 2018. Mitigating unwanted biases with adversarial learning. In Conference on Artificial Intelligence, Ethics and Society.
- Zhang et al. (2017) Sheng Zhang, Rachel Rudinger, Kevin Duh, and Benjamin Van Durme. 2017. Ordinal Common-sense Inference. Transactions of the Association for Computational Linguistics, 5:379–395.
- Zhao et al. (2017) Jieyu Zhao, Tianlu Wang, Mark Yatskar, Vicente Ordonez, and Kai-Wei Chang. 2017. Men also like shopping: Reducing gender bias amplification using corpus-level constraints. In Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, pages 2979–2989.
- Zhu et al. (2015) Yukun Zhu, Ryan Kiros, Richard Zemel, Ruslan Salakhutdinov, Raquel Urtasun, Antonio Torralba, and Sanja Fidler. 2015. Aligning books and movies: Towards story-like visual explanations by watching movies and reading books. In arXiv preprint arXiv:1506.06724.
- Zweig and Burges (2011) Geoffrey Zweig and Christopher JC Burges. 2011. The microsoft research sentence completion challenge. Technical report, Citeseer.