\documentclass{beamer}

\usepackage{beamerthemeblue}
\useinnertheme{circles}
\definecolor{pink}{rgb}{1, 0.2, 0.4}
\definecolor{blue}{rgb}{0,0.6,0.7}

\setbeamertemplate{navigation symbols}{} 

\title{Language Acquisition}
\author{Andreas van Cranenburgh\\ Arjan Nusselder\\ Nadya Peek\\ Carsten van Weelden\newline \newline
Universiteit van Amsterdam}
\date{\today}

\renewcommand{\emph}[1]{\textcolor{pink}{#1}}

\begin{document}

\frame{\titlepage}

\frame
{
\frametitle{Overview}
\begin{itemize}
\item Background
\item Hypothesis
\item Simulation overview
\item Schedule
\item Corpus Annotation
\item One Word Implementation
\item Two Word Specification
\item Two Word Implementation and Future
\end{itemize}
}


\frame
{
\frametitle{Background}
How do we acquire language?
\begin{itemize}
\item Competence vs. Performance
	\begin{itemize}
	\item \emph{Innate grammar} (Chomsky et. al.)
	\item \emph{Analogy-making} process (Scha et. al.)
	\end{itemize}
\end{itemize}

Innate grammar does not account for many phenomena such as the \emph{comprehension of ungrammatical sentences}.  Innate (generative) grammars do however produce \emph{grammatically sound incomprehensible constructions} such as \textit{I told him to tell him to tell him to tell him to say}.
}

\frame
{
\frametitle{Hypothesis}
If a child learns language by basing new grammatical structures on analogy with structures he has seen before,

\emph{How does a child acquire his first corpus?}

\bigskip

Stages of Language Acquisition
	\begin{itemize}
	\item \emph{Pre-linguistic stage:} categorizing specific situations into abstract situations.
	\item \emph{One-word stage:} associations of words with instances of Humans, Objects, Actions, Locations, etc.
	\item \emph{Two-word stage:} formation of first grammatical constructions.
	\item \emph{Multi-word stage:} formation of three+ word sentences.
	\end{itemize}
}

\frame
{
\frametitle{Cycle overview}
\begin{center}
	\includegraphics[width=1.00\textwidth]{overview}
\end{center}
}


\frame
{
\frametitle{Schedule}
Until now, we have remained on schedule.  We foresee that annotating the corpus may turn out to be more work than we hoped.  Nevertheless we expect to be able to remain on schedule.
\medskip

\begin{tabular}{|r|l|l|}
\hline
deadline&activity&remarks\\
\hline
Week 1&Background research&done\\
&Planning&done\\
\hline
Week 2&Implementation One-word stage& done\\
&Specification Two-word stage& done\\
\hline
Week 3&Implementation Two-word Stage&\\
&Annotating Corpus&\\
\hline
Week 4&Debugging&\\
&Report and Final presentation&\\
\hline
\end{tabular}

}

\frame
{
\frametitle{Corpus Annotation}
To be able to simulate a child extracting structure from its surroundings, we need a corpus which simulates a child's interactions with its surroundings.
\bigskip

\emph{CHILDES} (Child Language Data Exchange System) is an online corpus of child-adult conversations.

\bigskip

*FAT:	$<$show me$>$ [/] show me a few things in this picture .\\
*FAT:	what's happening here ?\\
*CHI:	happening there .\\
*CHI:	gettin(g) on the other side .\\
\bigskip

The corpus should be annotated with more complete situational descriptions, the \emph{frames}.  Frames should have rolls and contain an \emph{ID}, \emph{abstractions}, \emph{other frames} and \emph{properties}.

}


\frame
{
\frametitle{One-word Implementation}
\begin{itemize}
\item {\em language}: Python
\item {\em datastructure}: XML
\item {\em scoring}: todo
	\begin{itemize}
	\item associate frames with words using statistics
	\item cut of words like ``the,''  ``that'' etc.
	\end{itemize}
\end{itemize}
}

\frame
{\frametitle{oneword.py example output}
\tt{{\small
\$ python oneword.py \\
Talk to me: {\bf throw the block} \\
 throw \\
MEANING: \\
$\ \ \ \ $                FRAME: who \\
$\ \ \ \ $ $\ \ \ \ $                ABSTR: object:human \\
$\ \ \ \ $ $\ \ \ \ $                ID: child \\
$\ \ \ \ $ $\ \ \ \ $                PROP: family = yes \\
------------------------------------------------------------------------------- \\
the \\
MEANING: \\
$\ \ \ \ $        PROP: family = yes \\
------------------------------------------------------------------------------- \\
block \\
MEANING: \\
$\ \ \ \ $        PROP: shape = square \\
}
}}
\frame
{
\frametitle{Two-word Specification \\ Overview}
\begin{center}
	\includegraphics[width=1.00\textwidth]{2word}
\end{center}
}

\frame
{
\frametitle{Two-word Specification Cont. \\ Algorithm specification}
\emph{Analyze:} receives a situational interpretation with an utterance.   It searches for the most probable \emph{meaning} compatible with the SI.  It then searches for a \emph{linguistic abstraction}.
\begin{itemize}
\item
\emph{Reinforce:} if an appropriate LA is found, its occurrence is reinforced in the corpus.
\item
\emph{Hypothesize:} if no appropriate LA exists, we must create a new one.  
\end{itemize}
}

\frame
{
\frametitle{Two-word Specification Cont. \\ Creating a new LA}
The Hypothesize part of the 2-word stage is a little tricky.  However once a child \textit{expects} to learn, he will have more ease in making more Linguistic Abstractions.

\bigskip

Hypothesize (Utterance, Word-Meaning-List):
\begin{itemize}
\item
Find \emph{connected words}, where one meaning is a subframe of another.
\item
Make an \emph{abstraction} of the subframe. (e.g. balls can be red, yellow or blue)
\item
Add \emph{word-order}.
\end{itemize}
Output: [throw BEFORE Var]



}



\frame
{
\frametitle{Two-word Implementation}
Next week we will work on annotating the corpus and implementing the two-word stage.

\bigskip

The two-word specifications have already been written in pseudo code, so we hope we can implement the two-word stage in the third week, and will have time to run through multiple corpora for both the one- and two-word stage in the fourth week.

\bigskip

This of course will require a lot of corpus annotation.

}




\end{document}