SEMINAR

Jenny Myrendal (Ph.D. student in linguistics) and I have received funding for the Web Science project "Ett (data-)lingvistiskt perspektiv på webben". Since Web Science (see http://webscience.org) is an area where CLT wants to contribute (all according to the CLT Strategic Plan), I will give an overview of our project plans and also (with your help!) try to answer the question: What does linguistics and computational linguistics have to to with the Web?

SEMINAR

The presentation is a summary of my thesis, to be finished early in 2012. It explores parallel treebanks, texts and their translation which have been syntactically annotated and where corresponding parts are linked through alignment. Building parallel treebanks is a time-consuming and labour-intensive task. Additionally, for the treebank to be really useful, it needs to be large, containing millions of words. However, tools and/or annotated data are generally only available for a number of the largest languages in the world, and automatic annotation (as well as human annotation) induces errors, and we still need to check and manually correct the annotation, for the data to be useful.

The main question explored in this thesis is how to create a good quality parallel treebank. This is done in two separate ways. First, we describe how the Smultron parallel treebank was created. Second, we explore augmentation of automatically created annotation, in particular methods of projecting knowledge from multiple languages, known as multilingual annotation projection.

SEMINAR

Abstract:
I am developing a pedagogical tool for children with communication disabilities. The tool will consist of a computer system which gives support for language learning and language training. The system is graphical, where the child can point to words or phrases to get explanations in the form of symbols, written text, or spoken words. The child can also reformulate the sentences, by moving words to new places or change the inflection of words. The system will then automatically reformulate the sentence, so that it will always be grammatically correct. The tool is intended to stimulate - in a playful way - the child's will to explore the possibilities of language.

In this talk I will report on the current status of the tool, and the underlying ideas. If you are lucky, I might even show some pseudo-mathematical theories about tree similarity in type-theoretical grammars...

If you are even luckier, I might give you more awesome animations in the style of my legendary CLT-workshop-discussion-introduction!

SEMINAR

In this presentation, a dynamic semantic approach to subsymbolic perceptual aspects of meaning is presented. We show how a simple classifier of spatial information based on the Perceptron can be cast in TTR (Type Theory with Records). Furthermore, we show how subsymbolic aspects of meaning can be updated as a result of observing language use in interaction, thereby enabling fine-grained semantic plasticity and semantic coordination.

SEMINAR

Abstract:

Software models is one way of bridging the gap between the code and intentions behind a system. Over the years there has been a number of different modelling languages and frameworks for using them. One such language is Executable and Translatable UML (xtUML) which is one way of enabling a framework called Model-Driven Architecture (MDA). A core concept within MDA and xtUML is model transformations. This seminar will show how model transformations can be used for Natural Language Generation as a means for model validation.

SEMINAR

A dry run of three talks to be presented at the SemDial workshop in Los Angeles.

http://projects.ict.usc.edu/nld/semdial2011/

1. Kristina Lundholm Fors & Jessica Villing: "Reducing cognitive load in in-vehicle dialogue system interaction"

Abstract:

In-vehicle dialogue systems need to be able to adapt to the cognitive load of the user, and, when possible, reduce cognitive load. To accomplish this, we need to know how humans act while driving and talking to a passenger, and find out if there are dialogue strategies that can be used to minimize cognitive load. In this study, we have analyzed human-human in-vehicle dialogues, focusing on pauses and adjacency pairs. Our results show that when the driver is experiencing high cognitive load, the passenger’s median pause times increase. We also found that, when switching to another domain and/or topic, both driver and passenger try to avoid interrupting an adjacency pair. This suggests that a dialogue system could help lower the user’s cognitive load by in- creasing pause lengths within turns, and plan system utterances in order to avoid switching task within an adjacency pair.

2. Ellen Breitholz & Robin Cooper: "Enthymemes as Rhetorical Resources"

Abstract:

In this paper we propose that Aristotelian enthymemes play a role in the resources available to dialogue participants.  We take as our point of departure the idea that every individual has a set of linguistic resources that are formed and reformed through interaction with other individuals and context.

We regard enthymemes as dependent record types, functions which map contexts modelled as records, corresponding to the premises of the enthymeme, to a record type which models a proposition corresponding to the conclusion of the enthymeme. The advantage of using record types is that they give us semantic objects corresponding to enthymemes (as opposed to textual objects such as inference rules) and a straightforward way of generalizing, restricting and combining enthymemes thereby giving a theory of how agents can expand and reform their rhetorical resources on the basis of experience. In this paper we propose that Aristotelian enthymemes play a role in the resources available to dialogue participants.  We take as our point of departure the idea that every individual has a set of linguistic resources that are formed and reformed through interaction with other individuals and context.
 
We regard enthymemes as dependent record types, functions which map contexts modelled as records, corresponding to the premises of the enthymeme, to a record type which models a proposition corresponding to the conclusion of the enthymeme. The advantage of using record types is that they give us semantic objects corresponding to enthymemes (as opposed to textual objects such as inference rules) and a straightforward way of generalizing, restricting and combining enthymemes thereby giving a theory of how agents can expand and reform their rhetorical resources on the basis of experience.

3. Robin Cooper and Jonathan Ginzburg: "Negation in Dialogue"

Abstract:

We consider the nature of negation in dialogue as revealed by semantic phenomena such as negative dialogue particles, psycholinguistic experimentation, and dialogue corpora. We examine alternative accounts of negation that can be used in TTR (Type Theory with Records), and conclude that an alternatives-based account which relates to the psychological notion of negation in simulation semantics is most appropriate. We show how this account relates to questions under discussion, dialogical relevance, and metalinguistic negation.

SEMINAR

Hans Leiß (LMU, München)

Abstract:
A type reconstruction algorithm for a fragment of natural language is presented, based on Hindley's algorithm for simple types plus structural subtyping and overloading of constants. We extend Montague's PTQ-fragment of English by plural noun phrases (which may have several types) and overloaded verbs to allow for distributed and non-distributed readings of noun phrases and verb arguments. We demonstrate how type reconstruction can select suitable meanings of subject noun phrases depending on the meaning of verb phrases, thereby handling some violations of Frege's compositionality principle.

http://www.cis.uni-muenchen.de/~leiss/

SEMINAR

SEMINAR

During a 7.5hp project I've been working on a pilot project, evaluating and extending the existing GF grammar for Swedish. By using a large lexicon ported from SALDO and an extra GF module we hope to eventually implement a substantial parser for Swedish. This should be able to parse all of Talbanken, a treebank consisting of over 6000 sentences.

In the course of the project, grammatical constructs have been added to the GF grammar, and some improvements have been made to the lexicon. Also, an interactive tool for lexicon acquisition of Swedish verbs has been implemented, which builds GF lexicons by guessing the inflection paradigm for a given word. I will continue the work in my Master Project, and at the presentation I will discuss my results and the directions for future work that have been identified.

SEMINAR

Broadly speaking, "genre" is a classification concept. A genre is a recurring and recognized pattern of communication that has a specific name. The web hosts many recognised genres, such as FAQs, press releases, product descriptions, instructions, guides, e-magazines, blogs, professional profiles, how-tos, web ads and reviews. Each of these genres serves a number of communicative and social purposes and carries additional contextual information that helps the reader interpret the content. Can web genres be identified and detected automatically? Which computational models have been tried out so far in automatic genre identification research? How well do they perform? In this talk, I will present and discuss the latest findings in automatic genre identification and suggest viable future directions.

Web:

Academic: http://sites.google.com/site/marinasantiniacademicsite/

Artificial Solutions: http://www.artificial-solutions.com/