Guido's feedback on the summary sections

chapters/2_background.tex

@@ -1615,13 +1615,13 @@ It tries to reduce the number of Boolean \variables{} and tries to reduce the nu
 
 \section{Summary}
 
-This chapter summarized the key knowledge to understand the \gls{rewriting} \minizinc{} \instances{} to \glspl{slv-mod}.
+This chapter summarized gave an overview of the background knowledge required to understand the following technical chapters.
 It introduced the practice of \constraints{} modelling, and the syntax of the \minizinc{} language.
 We also compared \minizinc{} to other \cmls{} and found many similarities.
-This indicates that the research presented in this thesis could be used for these languages as well.
+This indicates that the research presented in this thesis could be applied to these languages as well.
 By using \cmls{}, a modeller can easily express a problem for a variety of \solver{} programs.
 We gave a brief overview of the main methods used by these \solvers{} and their problem formats, to which a \cmodel{} must be rewritten.
 Finally, we discussed the \gls{rewriting} process central to \cmls{} in more detail, focusing on the \gls{rewriting} conducted by the current implementation of \minizinc{}.
 
-The next chapter is the first of the three main empirical chapters of this thesis.
-It presents a more efficient architecture to perform the \gls{rewriting} from \minizinc{} \instances{} to solver models, based on a set of formalized rewriting rules that allows better reasoning about functional \constraints{}.
+The next chapter is the first of the three main technical chapters of this thesis.
+It presents a more efficient architecture to perform the \gls{rewriting} from \minizinc{} \instances{} to solver models, based on a set of formalized rewriting rules that support better reasoning about functional \constraints{}.

chapters/3_rewriting.tex

@@ -851,7 +851,7 @@ We are convinced, however, that we can get closer to its performance given the r
 \section{Summary}%
 \label{sec:rew-summary}
 
-This chapter presented a new architecture for \gls{rewriting} of \minizinc{} \instances{} to \glspl{slv-mod}.
+This chapter presented a new architecture for \gls{rewriting} \minizinc{} \instances{} to \glspl{slv-mod}.
 We introduced the intermediate languages \microzinc{} and \nanozinc{}, which are central to the \gls{rewriting} architecture.
 \minizinc{} is transformed into a set of \microzinc{} definitions and a \nanozinc{} program.
 Then, a \nanozinc{} program is partially evaluated using \microzinc{} definitions.

chapters/4_half_reif.tex

@@ -1192,14 +1192,14 @@ That the effect is so positive is surprising since its \gls{rewriting} statistic
 \section{Summary}
 \label{sec:half-summary}
 
-The usage of \gls{reified} \constraints{} can be detrimental to \solver{} and the \gls{rewriting} process should use non-\gls{reified} or \gls{half-reif} \constraints{} when possible.
+The usage of \gls{reified} \constraints{} can be detrimental to \solver{} performance and the \gls{rewriting} process should use non-\gls{reified} or \gls{half-reif} \constraints{} when possible.
 This chapter presented a framework to reason about \gls{reif} during the \gls{rewriting} process.
 It provided a \microzinc{} analysis that associates each expression with a certain context.
 Depending on the context, we then know whether a \constraint{} can remain non-\gls{reified}, can be \gls{half-reified}, or must be \gls{reified}.
 Notably, the best context cannot always be determined without a complete \gls{parameter-assignment} and \gls{propagation}.
-We showed how this can be corrected through the definitions of the \gls{reif} predicate.
+We showed how the best context for a \constraint{} can still always be used by introducing an additional check of the \gls{cvar}.
 Finally, we adapted the simplification techniques used during \gls{rewriting} to the use of \gls{half-reif}.
-We corrected \gls{cse} to take the context of an expression into account, and we introduced a new simplification technique, \gls{chain-compression}, to eliminate \glspl{implication-chain} introduced by \gls{half-reif}.
+We showed how \gls{cse} can be modified to take the context of an expression into account, and we introduced a new simplification technique, \gls{chain-compression}, to eliminate \glspl{implication-chain} introduced by \gls{half-reif}.
 
 To make full use of \gls{half-reif}, we extended \gls{chuffed} with two \glspl{propagator} for \gls{half-reif} \constraints{} that were shown to be effective for certain models.
 We also extended the \minizinc{} \gls{linearization} and \gls{booleanization} libraries to make use of \gls{half-reif} in their \glspl{decomp}.

chapters/5_incremental.tex

@@ -842,6 +842,6 @@ When required, however, the architecture presented in this thesis can efficientl
 The architecture provides an incremental interface that allows the modeller to add and later retract \variables{} and \constraints{} in chronological order.
 Internally, this \gls{rewriting} process is extended to maintain a \gls{trail} to be able to retract \constraints{} and minimize the portion of the \instance{} to be rewritten again.
 \Gls{incremental-rewriting} is shown to save a significant amount of time, compared to repeatedly \gls{rewriting} the full \instance{}.
-We also discussed how incremental changing \gls{slv-mod} can be communicated to the \solver{} at different levels, but we could not show a clear benefit of the incremental communication.
+We also discussed how incremental changes to \glspl{slv-mod} can be communicated to the \solver{} at different levels, but we could not show a clear benefit of the incremental communication.
 
 The next and final chapter presents the conclusions of this thesis, which reiterate the discoveries and contributions of this research to theory and practice, comment on the scope and limitations of the presented architecture, and present further avenues for research in this area.