Mark Alan Finlayson

Two-dimensional spatial-phase-locked electron-beam lithography via sparse sampling

We report a new mode of spatial-phase-locked electron-beam lithography based on alignment of each e-beam deflection field to a fiducial grid on the substrate. Before exposing the pattern in a given field, the fiducial grid is sparsely sampled with the electron beam at a subexposure dose. These samples form a two-dimensional moire pattern that is analyzed to calculate field shift, scale, rotation, nonorthogonality, and trapezoidal distortion. Experimental verification of the approach was carried out with a scintillating fiducial grid quenched by interference lithography. Despite a poor signal-to-noise ratio, we achieved sub-beamstep field-stitching and pattern-placement accuracy.

Computational Models of Narrative: Review of a Workshop

On October 8-10, 2009 an interdisciplinary group met at the Wylie Center in Beverley, Massachusetts to evaluate the state of the art in the computational modeling of narrative. Three important findings emerged: (1) current work in computational modeling is described by three different levels of representation; (2) there is a paucity of studies at the highest, most abstract level aimed at inferring the meaning or message of the narrative; and (3) there is a need to establish a standard data bank of annotated narratives, analogous to the Penn Treebank.

Overview of Annotation Creation: Processes and Tools

Creating linguistic annotations requires more than just a reliable annotation scheme. Annotation can be a complex endeavour potentially involving many people, stages, and tools. This chapter outlines the process of creating end-to-end linguistic annotations, identifying specific tasks that researchers often perform. Because tool support is so central to achieving high quality, reusable annotations with low cost, the focus is on identifying capabilities that are necessary or useful for annotation tools, as well as common problems these tools present that reduce their utility. Although examples of specific tools are provided in many cases, this chapter concentrates more on abstract capabilities and problems because new tools appear continuously, while old tools disappear into disuse or disrepair. The two core capabilities tools must have are support for the chosen annotation scheme and the ability to work on the language under study. Additional capabilities are organized into three categories: those that are widely provided; those that often useful but found in only a few tools; and those that have as yet little or no available tool support.

Toward Semantic Search for the Biogeochemical Literature

Literature search is a vital step of every research project. Semantic literature search is an approach to article retrieval and ranking using concepts rather than keywords, in an attempt to address the well-known deficiencies of keyword-based search, namely, (1) retrieval of an overwhelming number of results, (2) rankings that do not precisely reflect true relevance, and (3) the omission of relevant results because they do not contain the idiosyncratic keywords of the query. The difficulty of semantic search, however, is that it requires significant knowledge engineering, often in the form of conceptual ontologies tailored to a particular scientific domain. It also requires non-trivial tuning, in the form of domain-specific term and concepts weights. Here we present preliminary, work-in-progress results in the development of a semantic search system for the biogeochemical scientific literature. We report the following initial steps: first, one of the co-authors—a biogeochemistry expert—wrote a sample search query, and ranked the five most relevant articles that were returned for that query from a popular keyword-based search engine. We then hand annotated the five articles and the query with the Environmental Ontology (ENVO), an existing ontology for the domain. Critically, this pilot annotation revealed a number of missing concepts that we will add in future work. We then showed that a straightforward ontology distance metric between concepts in the search query and the five articles was sufficient to produce the expected ranking. We discuss the implications of these results, and outline next steps required produce a full-fledged semantic search system for the biogeochemistry scientific literature.

Narrative Constructions for the Organization of Self Experience: Proof of Concept via Embodied Robotics

It has been proposed that starting from meaning that the child derives directly from shared experience with others, adult narrative enriches this meaning and its structure, providing causal links between unseen intentional states and actions. This would require a means for representing meaning from experience—a situation model—and a mechanism that allows information to be extracted from sentences and mapped onto the situation model that has been derived from experience, thus enriching that representation. We present a hypothesis and theory concerning how the language processing infrastructure for grammatical constructions can naturally be extended to narrative constructions to provide a mechanism for using language to enrich meaning derived from physical experience. Toward this aim, the grammatical construction models are augmented with additional structures for representing relations between events across sentences. Simulation results demonstrate proof of concept for how the narrative construction model supports multiple successive levels of meaning creation which allows the system to learn about the intentionality of mental states, and argument substitution which allows extensions to metaphorical language and analogical problem solving. Cross-linguistic validity of the system is demonstrated in Japanese. The narrative construction model is then integrated into the cognitive system of a humanoid robot that provides the memory systems and world-interaction required for representing meaning in a situation model. In this context proof of concept is demonstrated for how the system enriches meaning in the situation model that has been directly derived from experience. In terms of links to empirical data, the model predicts strong usage based effects: that is, that the narrative constructions used by children will be highly correlated with those that they experience. It also relies on the notion of narrative or discourse function words. Both of these are validated in the experimental literature.

Automatic Identification of Narrative Diegesis and Point of View

The style of narrative news affects how it is interpreted and received by readers. Two key stylistic characteristics of narrative text are point of view and diegesis: respectively, whether the narrative recounts events personally or impersonally, and whether the narrator is involved in the events of the story. Although central to the interpretation and reception of news, and of narratives more generally, there has been no prior work on automatically identifying these two characteristics in text. We develop automatic classifiers for point of view and diegesis, and compare the performance of different feature sets for both. We built a goldstandard corpus where we double-annotated to substantial agreement (κ > 0.59) 270 English novels for point of view and diegesis. As might be expected, personal pronouns comprise the best features for point of view classification, achieving an average F1 of 0.928. For diegesis, the best features were personal pronouns and the occurrences of first person pronouns in the argument of verbs, achieving an average F1 of 0.898. We apply the classifier to nearly 40,000 news texts across five different corpora comprising multiple genres (including newswire, opinion, blog posts, and scientific press releases), and show that the point of view and diegesis correlates largely as expected with the nominal genre of the texts. We release the training data and the classifier for use by the community.

ProppML: A Complete Annotation Scheme for Proppian Morphologies

We give a preliminary description of ProppML, an annotation scheme designed to capture all the components of a Proppian-style morphological analysis of narratives. This work represents the first fully complete annotation scheme for Proppian morphologies, going beyond previous annotation schemes such as PftML, ProppOnto, Bod et al., and our own prior work. Using ProppML we have annotated Propps morphology on fifteen tales (18,862 words) drawn from his original corpus of Russian folktales. This is a significantly larger set of data than annotated in previous studies. This pilot corpus was constructed via double annotation by two highly trained annotators, whose annotations were then combined after discussion with a third highly trained adjudicator, resulting in gold standard data which is appropriate for training machine learning algorithms. Agreement measures calculated between both annotators show very good agreement (F_1>0.75, kappa>0.9 for functions; F_1>0.6 for moves; and F_1>0.8, kappa>0.6 for dramatis personae). This is the first robust demonstration of reliable annotation of Propps system.

Learning a Better Motif Index: Toward Automated Motif Extraction

Motifs are distinctive recurring elements found in folklore, and are used by folklorists to categorize and find tales across cultures and track the genetic relationships of tales over time. Motifs have significance beyond folklore as communicative devices found in news, literature, press releases, and propaganda that concisely imply a large constellation of culturally-relevant information. Until now, folklorists have only extracted motifs from narratives manually, and the conceptual structure of motifs has not been formally laid out. In this short paper we propose that it is possible to automate the extraction of both existing and new motifs from narratives using supervised learning techniques and thereby possible to learn a computational model of how folklorists determine motifs. Automatic extraction would enable the construction of a truly comprehensive motif index, which does not yet exist, as well as the automatic detection of motifs in cultural materials, opening up a new world of narrative information for analysis by anyone interested in narrative and culture. We outline an experimental design, and report on our efforts to produce a structured form of Thompsons motif index, as well as a development annotation of motifs in a small collection of Russian folklore. We propose several initial computational, supervised approaches, and describe several possible metrics of success. We describe lessons learned and difficulties encountered so far, and outline our plan going forward.

Comparing Extant Story Classifiers: Results & New Directions.

Having access to a large set of stories is a necessary first step for robust and wide-ranging computational narrative modeling; happily, language data - including stories - are increasingly available in electronic form. Unhappily, the process of automatically separating stories from other forms of written discourse is not straightforward, and has resulted in a data collection bottleneck. Therefore researchers have sought to develop reliable, robust automatic algorithms for identifying story text mixed with other non-story text. In this paper we report on the reimplementation and experimental comparison of the two approaches to this task: Gordons unigram classifier, and Cormans semantic triplet classifier. We cross-analyze their performance on both Gordons and Cormans corpora, and discuss similarities, differences, and gaps in the performance of these classifiers, and point the way forward to improving their approaches.

ProppLearner: Deeply Annotating a Corpus of Russian Folktales to Enable the Machine Learning of a Russian Formalist Theory

I describe the collection and deep annotation of the semantics of a corpus of Russian folktales. This corpus, which I call the ‘ProppLearner’ corpus, was assembled to provide data for an algorithm designed to learn Vladimir Propp’s morphology of Russian hero tales. The corpus is the most deeply annotated narrative corpus available at this time. The algorithm and learning results are described elsewhere; here, I provide detail on the layers of annotation and how they were chosen, novel layers of annotation required for successful learning, the selection of the texts for annotation, the annotation process itself, and the resulting inter-annotator agreement measures. In particular, the corpus comprised fifteen texts totaling 18,862 words. There were eighteen layers of annotation, five of which were developed specifically to support learning Propp’s morphology: referent attributes, context relationships, event valences, Propp’s ‘dramatis personae’, and Propp’s functions. All annotations were created by trained annotators with the Story Workbench annotation tool, following a double-annotation paradigm. I discuss lessons learned from this effort and what they mean for future digital humanities efforts when working with the semantics of natural language text.