http://www.era.lib.ed.ac.uk The DSpace digital repository system captures, stores, indexes, preserves, and distributes digital research material. Search the Channel search http://www.era.lib.ed.ac.uk/simple-search http://hdl.handle.net/1842/3317 Title: The St. Andrews psalter : transcription and critical study of Thomas Wode’s psalter.<br/><br/>Authors: Hutchison, Hilda<br/><br/>Abstract: The St-Andrews Psalter is the earliest harmonized Metrical Psaltor extant in Scotland. It was compiled by Thomas Wod, or Wode, Reader in the post-Reformation Church, and Vicar of St. Andrews from 1575 until his death. http://hdl.handle.net/1842/3316 Title: The production and degradation of trichloroacetic acid in soil: results from in situ soil column experiments<br/><br/>Authors: Heal, Mathew R; Dickey, Catherine A; Heal, Kate V; Stidson, Ruth T; Matucha, Miroslav; Cape, J Neil<br/><br/>Abstract: Previous work has indicated that the soil is important to understanding biogeochemical fluxes of trichloroacetic acid (TCA) in the rural environment, in forests in particular. Here, the hydrological and TCA fluxes through 22 in situ soil columns in a forest and moorland-covered catchment and an agricultural grassland field in Scotland were monitored every two weeks for several months either as controls or in TCA manipulation (artificial dosing) experiments. This was supplemented by laboratory experiments with radioactively-labelled TCA and with irradiated (sterilised) soil columns. Control in situ forest soil columns showed evidence of net export (i.e. in situ production) of TCA, consistent with a net soil TCA production inferred from forest-scale mass balance estimations. At the same time, there was also clear evidence of substantial in situ degradation within the soil (~70% on average) of applied TCA. The laboratory experiments showed that both the formation and degradation processes operate on time scales of up to a few days and appeared related more with biological rather than abiotic processes. Soil TCA activity was greater in more organic-rich soils, particularly within forests, and there was strong correlation between TCA and soil biomass carbon content. Overall it appears that TCA soil processes exemplify the substantial natural biogeochemical cycling of chlorine within soils, independent of any anthropogenic chlorine flux. http://hdl.handle.net/1842/3315 Title: Gaseous and particulate water-soluble organic and inorganic nitrogen in rural air in southern Scotland<br/><br/>Authors: Gonzalez Benitez, Juan M; Cape, J Neil; Heal, Mathew R<br/><br/>Abstract: Simultaneous daily measurements of water-soluble organic nitrogen (WSON), ammonium and nitrate were made between July and November 2008 at a rural location in south-east Scotland, using a ‘Cofer’ nebulizing sampler for the gas phase and collection on an open-face PTFE membrane for the particle phase. Average concentrations of NH3 were 82 ± 17 nmol N m-3 (error is s.d. of triplicate samples), while oxidised N concentrations in the gas phase (from trapping NO2 and HNO3) were smaller, at 2.6 ± 2.2 nmol N m-3, and gas-phase WSON concentrations were 18 ± 11 nmol N m-3. The estimated collection efficiency of the nebulizing samplers for the gas-phase was 88 (±8) % for NH3, 37 (±16) % for NO2 and 57 (±7) % for WSON; reported average concentrations have not been corrected for sampling efficiency. Concentrations in the particle phase were smaller, except for nitrate, at 21 ± 9, 10 ± 6 and 8 ± 9 nmol N m-3, respectively. The absence of correlation in either phase between WSON and either (NH3 + NH4+) or NO3- concentrations suggests atmospheric WSON has diverse sources. During wet days, concentrations of gas and particle-phase inorganic N were lower than on dry days, whereas the converse was true for WSON. These data represent the first reports of simultaneous measurements of gas and particle phase water-soluble nitrogen compounds in rural air on a daily basis, and show that WSON occurs in both phases, contributing 20-25% of the total water-soluble nitrogen in air, in good agreement with earlier data on the contribution of WSON to total dissolved N in rainfall in the UK. http://hdl.handle.net/1842/3314 Title: Tunnel ventilation effectiveness in fire scenarios<br/><br/>Authors: Colella, Francesco; Rein, Guillermo; Carvel, Ricky O; Torero, Jose L<br/><br/>Abstract: Throughout most of a tunnel network the ventilation behaviour may be approximated with a simple 1D flow model. However, there are some important - but relatively small - regions of the tunnel that require CFD analysis. The multi-scale model is theideal tool for such tunnel studies as it allows accurate flow field predictions in some locations, yet allows simplifications where highly detailed data are not required. It is shown that the accuracy of the multi-scalemodel is as high as the full CFD approach. The 100 times lower computational time is of great advantage because many ventilation scenarios can be explored and extensive sensitive parametric studies can be conducted.<br/><br/>Description: Article on the FS-World Magazine special Edition on Tunnel Safety (4th International Symposium on Tunnel Safety and Security Frankfurt, March, 2010) http://hdl.handle.net/1842/3313 Title: Choice of syntactic structure during language production: The production of unbounded dependencies<br/><br/>Authors: Huxley, Clare J<br/><br/>Abstract: During language production, conceptual messages are encoded into a targetlanguage and articulated. Existing models of language production assume severalstages of processing including a conceptual level, a level where lexical selection andsyntactic processing occurs and a level where morphological and phonologicalfeatures are added ready for production (e.g. Levelt et al., 1999). Previous researchhas considered how lexical and syntactic information could be stored via lemma(Kempen & Huijbers, 1983), syntactic nodes (Levelt at el., 1999) and combinatorialnodes (Pickering & Braingan, 1998), but little is understood about how syntacticstructures are selected. This thesis examines how constituent structures are selectedby investigating choice of structure in unbounded dependencies such as Which jugwith the red spots is the nun giving the monk? and how this is affected by factors suchas verb-subcategorisation preferences and global sentence structure complexity.A series of language production experiments investigate how globalstructure complexity and verb-subcategoricatisaion preferences affect choice ofsyntactic structure at the clause level in unbounded dependencies. A picturedescription task reveals an unusual preference for the dispreferred passive voicestructure as a result of global structural complexity. Sentence recall experimentsdemonstrate that both global structural complexity and verb-subcategorisationpreferences can affect choice of structure and that competition between these factorsdecides the final structure. Finally, syntactic priming experiments show thatprocessing mechanisms are shared between simple matrix clause structures andunbounded dependency clause structures, but that the influence of these sharedmechanisms vary between the different structure types. This could be attributed to amodal of processing where choice of structure is decided by competition betweenstructure representations which are influenced by different factors in differentglobal syntactic conditions.The results suggest that choice of syntactic structure is decided throughcompetition between possible structures. These possible structures may receive further activation or inhibition from other factors such as global structuralcomplexity or verb-subcategorisation preferences and thematic fit. Global structuralcomplexity may influence structure preferences through increased processing loador through attempts to integrate the clause structure with another global structure.Thematic role arguments may influence structure through a preference thatsyntactic roles fit with specified thematic roles. (e.g. experiencer as subject). Thismodel assumes parallel processing of possible structures and individual structureswithin a complex larger structure. It also assumes an incremental model ofprocessing which attempts to integrate structures as soon as possible. http://hdl.handle.net/1842/3312 Title: Modern Performance Practice and Aesthetics in Traditional Scottish Gaelic Singing<br/><br/>Authors: McPhee, Erin K<br/><br/>Abstract: Performance practice and aesthetics is an under-researched area in ScottishGaelic song. Previous studies have focused on text rather than context, and there is asignificant lack of information from the singers themselves, in their own words, as tohow they view their own tradition of singing and performance. The objective of thisstudy is to present an overview of modern performance practice and aesthetics inScottish Gaelic singing, based primarily on what singers believe to be important.The scope is not meant to be exhaustive; rather the purpose is to serve as apreliminary study of Gaelic singing and promote discussion in the topic. Located inthe field of ethnomusicology, it will consider not just the musical sounds produced,but the performance as a whole, and consequently the research is divided into thefour topics of aesthetics; singing styles and vocal techniques; performance contexts;and repertoire selection and transmission. The data were collected primarily throughfieldwork, combining interviews with fifteen Gaelic singers of different ages andexperience in Gaelic song with participant-observation at céilidhs, formal concerts,and Gaelic singing classes. The information gathered is revealing, indicating that,although singers have a clear sense of what is important to them in their singing,there is no universal set of criteria with which every singer agrees. Certain aspects ofsinging might inspire very definite opinions in some singers, and ambivalence inothers. But one prominent theme to have emerged, that appears to pervadethroughout modern Gaelic singing, is the juxtaposition of preservation versusinnovation. This manifests itself in many different ways, such as the contrastingcontexts of céilidh and concert hall; the contradictory opinions of how much interpretation is acceptable; whether or not singers choose songs based on thetraditional criterion of the poetry, or for the melody; whether or not accompanimentor choral singing has a place in traditional Gaelic song; and whether or not theirrepertoire selection is affected by the taste of modern, often non-Gaelic-speaking,audiences. All of this certainly reflects the state of change in which the world ofGaelic singing finds itself, with more external influences than ever before, and facingthe question of whether or not it can survive if it is not relevant for modern singersand audiences. http://hdl.handle.net/1842/3311 Title: Disfluencies affect language comprehension: evidence from event-related potentials and recognition memory<br/><br/>Authors: MacGregor, Lucy J<br/><br/>Abstract: Everyday speech is littered with disfluencies such as filled pauses, silent pauses,repetitions and repairs which reflect a speaker’s language production difficulties.But what are the effects on language comprehension?This thesis took a novel approach to the study of disfluencies by combining aninvestigation of the immediate effects on language processing with an investigationof the longer-term effects for the representation of language in memory. A series ofexperiments is reported which reflects the first attempt at a systematic investigationof the effects of different types of disfluencies on language comprehension.The experiments focused on the effects of three types of disfluencies—ers, silentpauses, and repetitions—on the comprehension of subsequent words. Critical wordswere either straightforward continuations of the pre-interrupted speech or a repairword which corrected the pre-interrupted speech. In addition, the effects that occurwhen er, repetition, and repair disfluencies themselves are processed, were assessed.ERPs showed that the N400 effect elicited in response to contextually unpredictablecompared to predictable words was attenuated by the presence of a pre-target erreflecting a reduction in the standard difference where unpredictable words are moredifficult to integrate into their contexts. This finding suggests that ers may reducethe extent to which listeners make predictions about upcoming words. In addition, words preceded by an er were more likely to be correctly recognised in a subsequentmemory test. These findings demonstrate a longer-term consequence for representationwhich may reflect heightened attention during processing. Silent pauses did notaffect the N400 but there was some indication of an effect on recognition memory.Repetition disfluencies did not affect the N400 or recognition memory. These findingsdemonstrate the importance of the nature of the disruption to speech. For alltypes of disfluent utterances, unpredictable words elicited a Late Positive Complex(LPC), possibly reflecting processes associated with memory retrieval and controlas listeners attempted to resume structural fluency after any interruption.Ers themselves elicited standard attention-related ERP effects: the Mismatch Negativity(MMN) and P300 effects, supporting the possibility that ers heighten attention.Repetition disfluencies elicited a right posterior positivity, reflecting detectionof the disfluency and possibly syntactic reanalysis. Repair disfluencies elicited anearly frontal negativity, possibly related to the detection of a word category violation,and a P600 effect, reflecting syntactic reanalysis. The presence of an erpreceding the repair eliminated the early negativity, but had no effect on the P600suggesting that ers may prepare listeners for the possibility of an upcoming repair,but that they do not reduce the difficulty associated with reanalysis.Taken together, the results from the studies reported in the thesis support an accountof disfluency processing which incorporates both prediction and attention http://hdl.handle.net/1842/3310 Title: Using Interior Point Methods for Large-scale Support Vector Machine training<br/><br/>Authors: Woodsend, Kristian<br/><br/>Abstract: Support Vector Machines (SVMs) are powerful machine learning techniques for classificationand regression, but the training stage involves a convex quadratic optimization programthat is most often computationally expensive. Traditionally, active-set methods have beenused rather than interior point methods, due to the Hessian in the standard dual formulationbeing completely dense. But as active-set methods are essentially sequential, they may notbe adequate for machine learning challenges of the future. Additionally, training time may belimited, or data may grow so large that cluster-computing approaches need to be considered.Interior point methods have the potential to answer these concerns directly. They scaleefficiently, they can provide good early approximations, and they are suitable for paralleland multi-core environments. To apply them to SVM training, it is necessary to addressdirectly the most computationally expensive aspect of the algorithm. We therefore present anexact reformulation of the standard linear SVM training optimization problem that exploitsseparability of terms in the objective. By so doing, per-iteration computational complexityis reduced from O(n3) to O(n). We show how this reformulation can be applied to manymachine learning problems in the SVM family.Implementation issues relating to specializing the algorithm are explored through extensivenumerical experiments. They show that the performance of our algorithm for large denseor noisy data sets is consistent and highly competitive, and in some cases can out perform allother approaches by a large margin. Unlike active set methods, performance is largely unaffectedby noisy data. We also show how, by exploiting the block structure of the augmentedsystem matrix, a hybrid MPI/Open MP implementation of the algorithm enables data andlinear algebra computations to be efficiently partitioned amongst parallel processing nodesin a clustered computing environment.The applicability of our technique is extended to nonlinear SVMs by low-rank approximationof the kernel matrix. We develop a heuristic designed to represent clusters using asmall number of features. Additionally, an early approximation scheme reduces the number of samples that need to be considered. Both elements improve the computational efficiencyof the training phase.Taken as a whole, this thesis shows that with suitable problem formulation and efficientimplementation techniques, interior point methods are a viable optimization technology toapply to large-scale SVM training, and are able to provide state-of-the-art performance. http://hdl.handle.net/1842/3309 Title: Evaluating Information Presentation Strategies for Spoken Dialogue Systems<br/><br/>Authors: Winterboer, Andi<br/><br/>Abstract: A common task for spoken dialogue systems (SDS) is to help users select a suitableoption (e.g., flight, hotel, restaurant) from the set of options available. When the numberof options is small, they can simply be presented sequentially. However, as thenumber of options increases, the system must have strategies for helping users browsethe space of available options.In this thesis, I compare two approaches to information presentation in SDS: (1)the summarize and refine (SR) approach (Polifroni et al., 2003; Polifroni, 2008) inwhich the summaries are generated by clustering the options based on attributes thatlead to the smallest number of clusters, and (2) the user-model based summarize andrefine (UMSR) approach (Demberg, 2005; Demberg and Moore, 2006) which employsa user model to cluster options based on attributes that are relevant to the user anduses coherence markers (e.g., connectives, discourse cues, adverbials) to highlight thetrade-offs among the presented items.Prior work has shown that users prefer approaches to information presentation thattake the user’s preferences into account (e.g., Komatani et al., 2003;Walker et al., 2004;Demberg and Moore, 2006). However, due to the complexity of building a workingend-to-end SDS, these studies employed an ”overhearer” evaluation methodology, inwhich participants read or listened to pre-prepared dialogues, thus limiting evaluationcriteria to users’ perceptions (e.g., informativeness, overview of options, and so on).In order to examine whether users prefer presentations based on UMSR when theywere actively interacting with a dialogue system, and to measure the effectiveness andefficiency of the two approaches, I compared them in a Wizard-of-Oz experiment. Ifound that in terms of both task success and dialogue efficiency the UMSR approachwas superior to the SR approach. In addition, I found that users also preferred presentationsbased on UMSR in the interactive mode.SDS are typically developed for situations in which the user’s hands and eyes arebusy. I hypothesized that the benefits of pointing out relationships among options (i.e.,trade-offs) in information presentation messages outweighs the costs of processingmore complex sentences. To test this hypothesis, I performed two dual task experimentscomparing the two approaches to information presentation in terms of theireffect on cognitive load. Again, participants performed better with presentations basedon the UMSR algorithm in terms of both dialogue efficiency and task success, and Ifound no detrimental effect on performance of the primary task.Finally, I hypothesized that one of the main reasons why UMSR is more efficientis because it uses coherence markers to highlight relations (e.g., trade-offs) betweenoptions and attributes. To test this hypothesis, I performed an eye-tracking experimentin which participants read presentations with and without these linguistic devices, andanswered evaluation and comparison questions to measure differences in item recall.In addition, I used reading times to examine comprehension differences between thetwo information presentation strategies. I found that the linguistic devices used inUMSR indeed facilitated item recall, with no penalty in terms of comprehension cost.Thus, in this thesis I showed that an approach to information presentation that employsa user model and uses linguistic devices such as coherence markers to highlighttrade-offs among the presented items improves information browsing. User studiesdemonstrated that this finding also applies to situations where users are performinganother demanding task simultaneously. http://hdl.handle.net/1842/3308 Title: Reforming designs: education and training in Scotland and Higher Still<br/><br/>Authors: Howieson, Cathy<br/><br/>Abstract: This thesis is concerned with the organisation of education and training systems and the extent to which their design may challenge or reinforce social inequalities. In modern societies, people’s life chances are inextricably linked to the education they are able to access and the knowledge and skills (typically manifested through formal qualifications) they acquire, thus how countries organise their education and training systems is of fundamental importance in determining the opportunities available to its citizens and to their life chances. The specific focus of the thesis is on the design and organisation of post compulsory education and training systems - a stage that represents a particular challenge for policy-makers - and within that, on how systems might conceptualise academic and vocational learning in more productive ways. Education systems are not context-free structures: the design of a nation’s education and training system provides a window onto its traditions, its social values and economic stance, and its current preoccupations and ambitions for itself. Thus the thesis uses the example of the Higher Still reform of post compulsory education and training in Scotland (from 1999 onwards) to reflect more generally on education and social inequalities in Scotland and to ask how we should understand the way in which Scotland has approached reform of its education and training provision. It seeks to explicate the reasons for the adoption of the Higher Still reform strategy, to identify the factors that determined its specific design and development and to reflect on how the particular reform strategy embodied by Higher Still relates to certain aspects of the Scottish context and its policy processes. The thesis then examines the institutional response to Higher Still and its impact on the opportunities available to young people. It locates the Higher Still example within the broader field of education policy, considering what the experience of the Higher Still reform reveals about the possibilities of re-designing an education and training system in ways which promote social equality and the scope for manoeuvre that policy-makers, in a specific national context, have in relation to system reform.