On L2 grammar and processing: the case of oblique relative clauses and the Null-Prep phenomenon

This dissertation investigates the acquisition of oblique relative clauses in L2 Spanish by English and Moroccan Arabic speakers in order to understand the role of previous linguistic knowledge and its interaction with Universal Grammar on the one hand, and the relationship between grammatical knowledge and its use in real-time, on the other hand. Three types of tasks were employed: an oral production task, an on-line self-paced grammaticality judgment task, and an on-line self-paced reading comprehension task. Results indicated that the acquisition of oblique relative clauses in Spanish is a problematic area for second language learners of intermediate proficiency in the language, regardless of their native language. In particular, this study has showed that, even when the learners’ native language shares the main properties of the L2, i.e., fronting of the obligatory preposition (Pied-Piping), there is still room for divergence, especially in production and timed grammatical intuitions. On the other hand, reaction time data have shown that L2 learners can and do converge at the level of sentence processing, showing exactly the same real-time effects for oblique relative clauses that native speakers had. Processing results demonstrated that native and non-native speakers alike are able to apply universal processing principles such as the Minimal Chain Principle (De Vincenzi, 1991) even when the L2 learners still have incomplete grammatical representations, a result that contradicts some of the predictions of the Shallow Structure Hypothesis (Clahsen & Felser, 2006). Results further suggest that the L2 processing and comprehension domains may be able to access some type of information that it is not yet available to other grammatical modules, probably because transfer of certain L1 properties occurs asymmetrically across linguistic domains. In addition, this study also explored the Null-Prep phenomenon in L2 Spanish, and proposed that Null-Prep is an interlanguage stage, fully available and accounted within UG, which intermediate L2 as well as first language learners go through in the development of pied-piping oblique relative clauses. It is hypothesized that this intermediate stage is the result of optionality of the obligatory preposition in the derivation, when it is not crucial for the meaning of the sentence, and when the DP is going to be in an A-bar position, so it can get default case. This optionality can be predicted by the Bottleneck Hypothesis (Slabakova, 2009c) if we consider that these prepositions are some sort of functional morphology. This study contributes to the field of SLA and L2 processing in various ways. First, it demonstrates that the grammatical representations may be dissociated from grammatical processing in the sense that L2 learners, unlike native speakers, can present unexpected asymmetries such as a convergent processing but divergent grammatical intuitions or production. This conclusion is only possible under the assumption of a modular language system. Finally, it contributes to the general debate of generative SLA since in argues for a fully UG-constrained interlanguage grammar.

Datasets, features, learning, and models in visual recognition

Visual recognition is a fundamental research topic in computer vision. This dissertation explores datasets, features, learning, and models used for visual recognition. In order to train visual models and evaluate different recognition algorithms, this dissertation develops an approach to collect object image datasets on web pages using an analysis of text around the image and of image appearance. This method exploits established online knowledge resources (Wikipedia pages for text; Flickr and Caltech data sets for images). The resources provide rich text and object appearance information. This dissertation describes results on two datasets. The first is Berg’s collection of 10 animal categories; on this dataset, we significantly outperform previous approaches. On an additional set of 5 categories, experimental results show the effectiveness of the method. Images are represented as features for visual recognition. This dissertation introduces a text-based image feature and demonstrates that it consistently improves performance on hard object classification problems. The feature is built using an auxiliary dataset of images annotated with tags, downloaded from the Internet. Image tags are noisy. The method obtains the text features of an unannotated image from the tags of its k-nearest neighbors in this auxiliary collection. A visual classifier presented with an object viewed under novel circumstances (say, a new viewing direction) must rely on its visual examples. This text feature may not change, because the auxiliary dataset likely contains a similar picture. While the tags associated with images are noisy, they are more stable when appearance changes. The performance of this feature is tested using PASCAL VOC 2006 and 2007 datasets. This feature performs well; it consistently improves the performance of visual object classifiers, and is particularly effective when the training dataset is small. With more and more collected training data, computational cost becomes a bottleneck, especially when training sophisticated classifiers such as kernelized SVM. This dissertation proposes a fast training algorithm called Stochastic Intersection Kernel Machine (SIKMA). This proposed training method will be useful for many vision problems, as it can produce a kernel classifier that is more accurate than a linear classifier, and can be trained on tens of thousands of examples in two minutes. It processes training examples one by one in a sequence, so memory cost is no longer the bottleneck to process large scale datasets. This dissertation applies this approach to train classifiers of Flickr groups with many group training examples. The resulting Flickr group prediction scores can be used to measure image similarity between two images. Experimental results on the Corel dataset and a PASCAL VOC dataset show the learned Flickr features perform better on image matching, retrieval, and classification than conventional visual features. Visual models are usually trained to best separate positive and negative training examples. However, when recognizing a large number of object categories, there may not be enough training examples for most objects, due to the intrinsic long-tailed distribution of objects in the real world. This dissertation proposes an approach to use comparative object similarity. The key insight is that, given a set of object categories which are similar and a set of categories which are dissimilar, a good object model should respond more strongly to examples from similar categories than to examples from dissimilar categories. This dissertation develops a regularized kernel machine algorithm to use this category dependent similarity regularization. Experiments on hundreds of categories show that our method can make significant improvement for categories with few or even no positive examples.