Samar Husain’s research while affiliated with Indian Institute of Technology Delhi and other places

Delaying the appearance of a verb in a noun-verb dependency tends to increase processing difficulty at the verb; one explanation for this locality effect is decay and/or interference of the noun in working memory. Surprisal, an expectation-based account, predicts that delaying the appearance of a verb either renders it no more predictable or more predictable, leading respectively to a prediction of no effect of distance or a facilitation. Recently, Husain et al. (2014) suggested that when the exact identity of the upcoming verb is predictable (strong predictability), increasing argument-verb distance leads to facilitation effects, which is consistent with surprisal; but when the exact identity of the upcoming verb is not predictable (weak predictability), locality effects are seen. We investigated Husain et al.'s proposal using Persian complex predicates (CPs), which consist of a non-verbal element—a noun in the current study—and a verb. In CPs, once the noun has been read, the exact identity of the verb is highly predictable (strong predictability); this was confirmed using a sentence completion study. In two self-paced reading (SPR) and two eye-tracking (ET) experiments, we delayed the appearance of the verb by interposing a relative clause (Experiments 1 and 3) or a long PP (Experiments 2 and 4). We also included a simple Noun-Verb predicate configuration with the same distance manipulation; here, the exact identity of the verb was not predictable (weak predictability). Thus, the design crossed Predictability Strength and Distance. We found that, consistent with surprisal, the verb in the strong predictability conditions was read faster than in the weak predictability conditions. Furthermore, greater verb-argument distance led to slower reading times; strong predictability did not neutralize or attenuate the locality effects. As regards the effect of distance on dependency resolution difficulty, these four experiments present evidence in favor of working memory accounts of argument-verb dependency resolution, and against the surprisal-based expectation account of Levy (2008). However, another expectation-based measure, entropy, which was computed using the offline sentence completion data, predicts reading times in Experiment 1 but not in the other experiments. Because participants tend to produce more ungrammatical continuations in the long-distance condition in Experiment 1, we suggest that forgetting due to memory overload leads to greater entropy at the verb.

This is the first attempt at characterizing reading difficulty in Hindi using naturally occurring sentences. We created the Potsdam-Allahabad Hindi Eyetracking Corpus by recording eye-movement data from 30 participants at the University of Allahabad, India. The target stimuli were 153 sentences selected from the beta version of the Hindi-Urdu treebank. We find that word- or low-level predictors (syllable length, unigram and bigram frequency) affect first-pass reading times, regression path duration, total reading time, and outgoing saccade length. An increase in syllable length results in longer fixations, and an increase in word unigram and bigram frequency leads to shorter fixations. Longer syllable length and higher frequency lead to longer outgoing saccades. We also find that two predictors of sentence comprehension difficulty, integration and storage cost, have an effect on reading difficulty. Integration cost (Gibson, 2000) was approximated by calculating the distance (in words) between a dependent and head; and storage cost (Gibson, 2000), which measures difficulty of maintaining predictions, was estimated by counting the number of predicted heads at each point in the sentence. We find that integration cost mainly affects outgoing saccade length, and storage cost affects total reading times and outgoing saccade length. Thus, word-level predictors have an effect in both early and late measures of reading time, while predictors of sentence comprehension difficulty tend to affect later measures. This is, to our knowledge, the first demonstration using eye-tracking that both integration and storage cost influence reading difficulty.

Processing cost is known to increase with dependency distance (Gibson 2000). However, the expectation-based account (Hale 2001, Levy 2008) predicts that delaying the appearance of a verb renders it more predictable and therefore easier to process. We tested the predictions of these two opposing accounts using complex predicates in Persian. One type of Complex predicate is a Noun-Verb configuration in which the verb is highly predictable given the noun. We delayed the appearance of the verb by interposing a relative clause (Expt 1, 42 subjects), or a single long PP (Expt 2, 43 subjects); the precritical region (the phrase before the verb) in both the short and long conditions was a short PP. Locality accounts such as Gibson (2000) predict a slowdown at the verb (real verb) due to increased Noun-Verb distance, whereas expectation accounts predict that distance should not adversely affect the processing time at the verb (because the conditional probability of the verb given the preceding context is close to 1-this was established with an offline sentence completion study). As a control, we included a simple predicate (Noun-Verb) configuration; the same distance manipulation was applied here as for complex predicates. In the control, locality accounts predict a slowdown in the long distance condition, but expectation accounts predict a speedup due to the increasing probability of the verb appearing given left context. Thus, we had a 2x2 design (high vs low predictability; short vs long distance). In Expt 1, we found a main effect of distance (t=4.24): reading time (RT) was longer in the long-distance conditions; a nested comparison showed that this effect was due to the low-predictable (simple predicate) conditions. In addition, both the high-predictable conditions were read faster than the low-predictable conditions (t=3.49). Expt 2, which had a long intervening PP, showed an even stronger main effect of distance (t=6.04) than in Expt 1: the RT in long conditions was slower than in short conditions; the locality effects were equally strong in the high and low predictable cases. As in Expt 1, we saw faster RTs in the high-predictable conditions. A combined analysis of the two experiments revealed a main effect of prediction (t=3.55) and a main of distance (t=4.30) as well as a marginal 3-way interaction between experiment, distance and prediction (t=-1.94). Thus, we find clear effects of locality in both experiments, and we also find evidence for expectation effects: the high-predictable verbs are read faster than the low-predictable verbs. The fact that we don't see facilitation with increased distance at the verb in spite of high predictability might be due to increased difficulty in prediction maintenance due to processing load. Recall that the locality effect in Expt 1 is driven only by low-predictable condition, while in Expt 2 both high and low are affected. In Expt 2, the intervener is a long, uninterrupted phrase whereas in Expt 1, the intervener consists of a short RC followed by a PP. Processing a single long intervening phrase may be harder than processing two different phrases, reminiscent of the sausage machine proposal of Frazier and Fodor (1978). The results suggest that complexity of intervening material is critical for prediction maintenance. Although we found evidence for both locality and expectation effects, a key prediction of the expectation account was not validated: delaying the appearance of a verb (predictable or not) did not facilitate processing.

Expectation-driven facilitation (Hale, 2001, Levy, 2008) and locality-driven retrieval difficulty (Gibson, 1998, 2000; Lewis & Vasishth, 2005) are widely recognized to be two critical factors in incremental sentence processing; there is accumulating evidence that both can influence processing difficulty. However, it is unclear whether and how expectations and memory interact. We first confirm a key prediction of the expectation account: a Hindi self-paced reading study shows that when an expectation for an upcoming part of speech is dashed, building a rarer structure consumes more processing time than building a less rare structure. This is a strong validation of the expectation-based account. In a second study, we show that when the expectation is strong, i.e., when a particular verb is predicted, strong facilitation effects are seen when the appearance of the verb is delayed; however, when expectation is weak, i.e., when only the part of speech ``verb'' is predicted but a particular verb is not predicted, the facilitation disappears and a tendency towards a locality effect is seen. The interaction seen between expectation strength and distance shows that strong expectations cancel locality effects, and that weak expectations allow locality effects to emerge.

Treebanks play an important role in the development of various natural language processing tools. Amongst other things, they provide crucial language-specific patterns that are exploited by various machine learning techniques. Quality control in any treebanking project is therefore extremely important. Manual validation of the treebank is one of the steps that is generally necessary to ensure good annotation quality. Needless to say, manual validation requires a lot of human time and effort. In this paper, we present an automatic approach which helps in detecting potential errors in a treebank. We use a dependency parser to detect such errors. By using this tool, validators can validate a treebank in less time and with reduced human effort.

The overall goal of our work is to build a dependency grammar-based human sen-tence processor for Hindi. As a first step towards this end, in this paper we present a dependency grammar that is motivated by psycholinguistic concerns. We describe the components of the grammar that have been automatically induced using a Hindi dependency treebank. We relate some as-pects of the grammar to relevant ideas in the psycholinguistics literature. In the pro-cess, we also extract statistics and pat-terns for phenomena that are interesting from a processing perspective. We fi-nally present an outline of a dependency grammar-based human sentence processor for Hindi.

We present two approaches (rule-based and statistical) for automatically annotating intra-chunk dependencies in Hindi. The intra-chunk dependencies are added to the dependency trees for Hindi which are already annotated with inter-chunk dependencies. Thus, the intra-chunk annotator finally provides a fully parsed dependency tree for a Hindi sentence. In this paper, we first describe the guidelines for marking intra-chunk dependency relations. Although the guidelines are for Hindi, they can easily be extended to other Indian languages. These guidelines are used for framing the rules in the rule-based approach. For the statistical approach, we use MaltParser, a data driven parser. A part of the ICON 2010 tools contest data for Hindi is used for training and testing the MaltParser. The same set is used for testing the rule-based approach.

We present a detailed error analysis of a transition-based dependency parser trained on a Hindi dependency treebank. Parser error analysis has not been systematically examined from the point of view of treebanking before and this work intends to contribute in this area. We address two main questions in this paper: Can the parsing of certain structures be made easier by using alternative analyses for these structures? Are there certain linguistic cues implicit (or missing) in the current treebank that can be made explicit (or added) in order to make the parsing of complex constructions easier? These questions will guide us in examining the potential benefits of parser error analysis during treebanking. Through our experiments and analysis we were able to shed light on the causes of errors and subsequently have been able to improve the performance of the parser.

In this paper we show how linguistic knowledge can be incorporated during graph based parsing. We use MSTParser and show that the use of a constraint graph, instead of a complete graph, to extract a spanning tree improves parsing accuracy. A constraint graph is formed by using linguistic knowledge of a constraint based parsing system. Through a series of experiments we formulate the optimal constraint graph that gives us the best accuracy. These experiments show that some of the previous MSTParser errors can be corrected consistently. It also shows the limitations of the proposed approach.

This paper introduces a work on identification of conjunct verbs in Hindi. The paper will first focus on investigating which noun-verb combination makes a conjunct verb in Hindi using a set of linguistic diagnostics. We will then see which of these diagnostics can be used as features in a MaxEnt based automatic identification tool. Finally we will use this tool to incorporate certain features in a graph based dependency parser and show an improvement over previous best Hindi parsing accuracy.