In this paper we explore the problem of biasing unsupervised models to favor sparsity. We extend the posterior regularization framework [8] to encourage the model to achieve posterior sparsity on the unlabeled training data. We apply this new method to learn first-order HMMs for unsupervised part-of-speech (POS) tagging, and show that HMMs learned this way consistently and significantly out-performs both EM-trained HMMs, and HMMs with a sparsity-inducing Dirichlet prior trained by variational EM. We evaluate these HMMs on three languages — English, Bulgarian and Portuguese — under four conditions. We find that our method always improves performance with respect to both baselines, while variational Bayes actually degrades performance in most cases. We increase accuracy with respect to EM by 2.5%-8.7% absolute and we see improvements even in a semisupervised condition where a limited dictionary is provided.

Language comprehension in humans is significantly constrained by memory, yet rapid, highly incremental, and capable of utilizing a wide range of contextual information to resolve ambiguity and form expectations about future input. In contrast, most of the leading psycholinguistic models and fielded algorithms for natural language parsing are non-incremental, have run time superlinear in input length, and/or enforce structural locality constraints on probabilistic dependencies between events. We present a new limited-memory model of sentence comprehension which involves an adaptation of the particle filter, a sequential Monte Carlo method, to the problem of incremental parsing. We show that this model can reproduce classic results in online sentence comprehension, and that it naturally provides the first rational account of an outstanding problem in psycholinguistics, in which the preferred alternative in a syntactic ambiguity seems to grow more attractive over time even in the absence of strong disambiguating information.

We develop the syntactic topic model (STM), a nonparametric Bayesian model of parsed documents. The STM generates words that are both thematically and syntactically constrained, which combines the semantic insights of topic models with the syntactic information available from parse trees. Each word of a sentence is generated by a distribution that combines document-specific topic weights and parse-tree-specific syntactic transitions. Words are assumed to be generated in an order that respects the parse tree. We derive an approximate posterior inference method based on variational methods for hierarchical Dirichlet processes, and we report qualitative and quantitative results on both synthetic data and hand-parsed documents.

This paper includes a reflection on the role of networks in the study of English language acquisition, as well as a collection of practical criteria to annotate free-speech corpora from children utterances. At the theoretical level, the main claim of this paper is that syntactic networks should be interpreted as the outcome of the use of the syntactic machinery. Thus, the intrinsic features of such machinery are not accessible directly from (known) network properties. Rather, what one can see are the global patterns of its use and, thus, a global view of the power and organization of the underlying grammar. Taking a look into more practical issues, the paper examines how to build a net from the projection of syntactic relations. Recall that, as opposed to adult grammars, early-child language has not a well-defined concept of structure. To overcome such difficulty, we develop a set of systematic criteria assuming constituency hierarchy and a grammar based on lexico-thematic relations. At the end, what we obtain is a well defined corpora annotation that enables us i) to perform statistics on the size of structures and ii) to build a network from syntactic relations over which we can perform the standard measures of complexity. We also provide a detailed example.

Intelligent Input Methods (IM) are essential for making text entries in many East Asian scripts, but their application to other languages has not been fully explored. This paper discusses how such tools can contribute to the deve lopment of computer processing of other oriental languages. We propose a design philosophy that regards IM as a text service platform, and treats the study of IM as a cross disciplinary subject from the perspectives of software engineering, human - computer interaction (HCI), and natural language processing (NLP). We discuss these three perspectives and indicate a number of possible future research directions.

In numerous domains in cognitive science it is often useful to have a source for randomly generated corpora. These corpora may serve as a foundation for artificial stimuli in a learning experiment (e.g., Ellefson & Christiansen, 2000), or as input into computational models (e.g., Christiansen & Dale, 2001). The following compact and general C program interprets a phrasestructure grammar specified in a text file. It follows parameters set at a Unix or Unix-based command-line and generates a corpus of random sentences from that grammar. The first and required input into the program is a file that contains a phrase-structure grammar description (see below).

In this paper, current dependency-based treebanks are introduced and analyzed. The methods used for building the resources, the annotation schemes applied, and the tools used (such as POS taggers, parsers and annotation software) are discussed.

Latent Semantic Analysis (LSA) is a widely used Information Retrieval method based on "bag-of-words" assumption. However, according to general conception, syntax plays a role in representing meaning of sentences. Thus, enhancing LSA with part-of-speech (POS) information to capture the context of word occurrences appears to be theoretically feasible extension. The approach is tested empirically on a automatic essay grading system using LSA for document similarity comparisons. A comparison on several POS-enhanced LSA models is reported. Our findings show that the addition of contextual information in the form of POS tags can raise the accuracy of the LSA-based scoring models up to 10.77 per cent.

DepAnn is an interactive annotation tool for dependency treebanks, providing both graphical and text-based annotation interfaces. The tool is aimed for semi-automatic creation of treebanks. It aids the manual inspection and correction of automatically created parses, making the annotation process faster and less error-prone. A novel feature of the tool is that it enables the user to view outputs from several parsers as the basis for creating the final tree to be saved to the treebank. DepAnn uses TIGER-XML, an XML-based general encoding format for both, representing the parser outputs and saving the annotated treebank. The tool includes an automatic consistency checker for sentence structures. In addition, the tool enables users to build structures manually, add comments on the annotations, modify the tagsets, and mark sentences for further revision.

Studies of different term extractors on a corpus of the biomedical domain revealed decreasing performances when applied to highly technical texts. The difficulty or impossibility of customising them to new domains is an additional limitation. In this paper, we propose to use external terminologies to influence generic linguistic data in order to augment the quality of the extraction. The tool we implemented exploits testified terms at different steps of the process: chunking, parsing and extraction of term candidates. Experiments reported here show that, using this method, more term candidates can be acquired with a higher level of reliability. We further describe the extraction process involving endogenous disambiguation implemented in the term extractor YaTeA.