and which papers employ which vocabularies, I
hope to trace the origin of these vocabularies as
well. And, if the vocabularies are specific enough,
then it may be that new vocabularies emerge
precisely where new ideas do. Therefore, to
determine the histo
context A B. This finally results in three different
values for x. As the illustrated store shows, x has
value 2, when features A and B are selected, value
1, if only A is selected, and 0, when none of the two
is selected. Assertions. To perform analysis
is not selected, which is shown in the store right to
the code fragment. VA Store A 0 2 x A B B 1 Figure
3.5: Variability propagation in blocks. Block
statements. On executing block statements it is
important to propagate variability of a whole block
to i
These approaches followed the spirit of the original
Chomskian revolution by focusing on representing
language as part of a formal mathematical
framework, notably certain variants of
mathematical logic. Moreover, they were
characterized by highly specific
have one block for their then branch and another
block for the else branch. Function declarations also
count as statements and include the function name,
a list of function parameters and the function body
as a block statement. Expressions can be numbers
selected, and value 3 otherwise. While loops. The
fact that conditions can be dependent on
configurations causes the problem that while loops
may have different numbers of iterations in
variability-aware interpretation. Therefore, while
loops are executed
which has also been used in the above example, is
the evaluation of expressions. Conceptually, the
interpreter takes a conditional expression and
returns a conditional value. In this case, every
alternative expression must be evaluated separately
in its o
Afterwards, Section 2.4.2 showed, how variability is
woven into ASTs when the code they represent is
enriched with variability. In the following, it is
illustrated how the AST nodes are implemented as
constructs of Scala. 1 abstract c l a s s Stmt 2 case
in more than one of these subfields. For example, in
recent years, members of the Machine Translation
sub-community have been influenced by work in
the Parsing sub-community. Fourth, there have
been a few studies on paradigm shifts in Natural
Language Pro
configurations. An example of early joining is shown
in Listing 3.1. In this code fragment, two functions
are declared: max(a,b) returns the maximum value
of a and b, while expensive(n) is a placeholder
representing any sort of CPU-intensive calculation.
particular revolution in the Computational
Linguistics community known as the Statistical
Revolution. In the span of just five years, the use of
statistics and probability in Computational
Linguistics went from fringe idea to dominant
paradigm. Who was re
the interpretation of two exemplary configurations
of the program and the variability-aware
interpretation. All program variables are tracked and
shown at five points of program execution. After the
first assignment (Line 2) all variables are shared
betwe
statement if assign block while assign | if | while |
block identifier = cfw_expr ; if ( cfw_expr ) block [else
block] while ( cfw_expr ) block cfw_ cfw_statement* expr
add add | sub | mul | div | grt | let | goe | leo |
and | or | eq | neq | neg | var |
down the street, a parsing program would
automatically determine that ran is the main verb,
that Pat is the subject, that down the street is a
prepositional phrase that modified ran, and
several other more low-level facts about the
sentence. During the 19
paradigms may be sufficient to identify them. How
then can one separate one vocabulary from
another? Most current approaches concern
themselves with latent topics, which are best
thought of as particular vocabularies shared across
papers [Hofmann, 1999].
cos(d1, d2) = d1 d2 |d1|d2| (2.2) , where |
d| = qP i C2 d,wi . By this measure, we have
similarities: cos(dA, dS) .35 cos(dA, dB) .32
cos(dS, dB) .45 2.1.2 Latent Semantic Analysis
However, both of these metrics have a substantial
problem. Ideally, I wou
were characterized by high-coverage but poor
accuracy. NLP practitioners, Gazdar reasons, realized
that these two communities could benefit from
each other. In particular, he points to signs that the
Parsing community (historically a part of the LOGIC
com
support high-level programming languages as Java,
C, C+, etc. in the future, this work presents a
prototype interpreter for initial insights on viability
and performance. Adapting the structures
presented in this work, to high-level programming
languages,
document as a mixture of several different
vocabularies: a biology paper may contains words
relating to statistical analysis, some kind of
experimental procedures, background about DNA,
and about photosynthesisto arbitrarily name
several possible vocabula
iteration the intermediate result is stored in variable
sum. Previously, x has been assigned (Lines 2- 5) and
affects the number of iterations, the while loop will
perform (Line 9). Above, it has been mentioned,
that the variabilityaware interpreter execu
more tractable computationally. 2.1.1
Understanding distance and similarity By appealing
to a mathematical representation of documents, we
can also consider how similar (or different) two
documents are. Intuitively, the more different two
document vectors
particular model: Latent Dirichlet Allocation (LDA;
Blei et al. [2003]). LDA is a fully generative version of
probabilistic Latent Semantic Analysis that models
(a version of) the process by which documents are
created. Instead of representing p(z|d), we
the uniform transition probability denoted by 1/|Cd
0| with the topicborrowing probability d 0d.
Chapter 5 Methodology This chapter seeks to
answer the questions outlined in the introduction:
who is most responsible for the Statistical
Revolution in Compu
novel model that directly models the progress of
normal science for determining how scientists
borrow ideas from their predecessors. Chapter Five
discusses the methodology used to analyze the
questions posed in the Introduction, while Chapter
Six presents
person. 1.2. TOPICS AND IDEAS 3 proportional to
the (square root of the) length, and not to the mass
at all. The depth of the incommensurability of
paradigms goes even further. Members of one
paradigm cannot logically argue for their paradigm
over another
WORK 3.1 Topic Dynamics A small but increasing
body of work has focused on topic dynamics. That
is, how do topics, and the words used to described
them, change over time? While it is possible to
analyze at least the question of topic prominence
post hoc,
thesis have a history rooted in the Vector Space
Model, in which one represents documents as
vectors of word counts [Manning et al., 2008]. For
example, consider three very short documents: 1.
Words are the physicians of a diseased mind.
Aeschylus 9 10 C
Dirichlet Allocations. At each epoch, the
distribution that topics and words are sampled from
changes. However, the Dynamic Topic Model
employs a normal distribution and a normalization
function to constrain the points to be on the
simplex. The model is s
may depend on the configuration, which implies
that possibly both branches must be visited. VA
Store A A 2 3 res Figure 3.3: Variability propagation
in if statements. The variability-aware interpreter
solves this problem by firstly checking, in which
conf
rigorous definition would help. Instead, based on
the Kuhn [1962] analysis from Chapter 1, it is
sufficient to discover when a scientist has borrowed
from another scientist. And for that, one needs a
way to characterize the symptoms of idea
transmission.