--- /dev/null
+\begin{enumerate}[label=\alph*.]
+ % 1a
+ \item
+ There are several differences between dialogues of native speakers
+ without a translator and with an intermediate translator.
+
+ When native speakers are aware of the fact that there is a translation
+ process going they will adapt the conversation style to be more
+ suitable for translation. This is because translation suffers from
+ common problems such as overlap between words and differences in
+ lexical structure that can cause confusion between the speakers.
+
+ For one the native speakers might use a lot more grounding to make sure
+ the translation was correct and the partner understands the thing that
+ has been said. Moreover, the grounding is probably a lot more
+ deliberate in the case of machine translation because a simple
+ backchannel like \emph{uhu} might not be enough to convince the partner
+ that the utterance was understood. This elaborate grounding might be
+ necessary since translation happens both ways and an utterance in
+ language \emph{A} translated to language \emph{B} and back to \emph{A}
+ could be different from the utterance that they started with. When two
+ native speakers converse the grounding can be very automatic and
+ simple. Speakers can use nuances in the grounding to determine whether
+ the listener understood the utterance and adapt thereupon.
+
+ Secondly, regarding conversation turn taking there is also likely a
+ change in conversation. Since intricate turn taking behaviour is a lot
+ more difficult to translate the turn-taking will likely be much more
+ concrete and structured. Because the turn-taking is so concrete it is
+ much harder to use references that span over turns.
+
+ Thirdly, the language use will be much more simple in the case of a
+ translator intervening. Lexical divergence can only be solved by a
+ translator when the context is explaining enough. When native speakers
+ converse the context might be implicit and subtle whereas in translator
+ the context must be concrete. Speakers will adapt to this and provide
+ more context, for example in the form of more adjectives. This remark
+ also includes the use of linguistic constructions that are very
+ difficult to translate such as metaphors and complex referencing.
+
+ Fourth and lastly, the use of conversational implicature will have to
+ be minimised since by mistranslations it might not be clear what the
+ implicature is. Moreover, the speakers are probably not sure what to
+ imply since the partner might come from a very different culture with a
+ very different language structure. For example when an English speaker
+ says \emph{a couple} it probably means \emph{two or more}. In languages
+ that also posses a dual case, such as Russian, next to the common
+ singular and plural this might be translated as \emph{exactly two}.
+ Because of errors of this kind the maxims proposed by Grice are extra
+ important.
+
+ % 1b
+ \item
+ The dialogue manager component should use the aforementioned techniques
+ to improve the understanding and clearness of the conversation.
+ It must use sophisticated grounding techniques such as \emph{explicit
+ confirmation} and \emph{rejection} when there is even a minimal amount
+ of doubt.
+
+ When the system is still not sure it can use rapid reprompting to get
+ the details clear, even when in the native language of the user it
+ might have been clear from other signals such as context.
+
+ This means that the dialogue manager probably should not produce a lot
+ of references, must expect very little implicature from the user and
+ must not use ambiguities.
+
+ When the user reprompts the system it should use a different, maybe a
+ little bit more illogical, construction to say the same. By doing this
+ the translation might be a bit better and therefore easier to
+ understand for the non-native speaker.
+
+ % 1c
+ \item
+ Evaluating dialogue systems can be done via multiple perspectives.
+
+ Expectations on task completion are pretty high. When an error occurs
+ the system can just reprompt. Moreover, when the user does not
+ understand an utterance it can also ask for a reprompt to the system
+ which will then hopefully reformulate the utterance.
+
+ Expectations on efficiency will probably be a lot worse than a system
+ without translation. There will be a lot more grounding, reprompting
+ and other clarification and confirmation techniques. All these
+ technique increase understanding at the cost of efficiency. Luckily in
+ an information providence system the conversations are often short and
+ therefore the overhead will not be as devastating.
+
+ Quality cost expectations will also not be as good as without
+ translation since due to all the problems mentioned above the system
+ has to use more recovery techniques which lower the quality.
+\end{enumerate}
--- /dev/null
+\begin{enumerate}[label=\alph*.]
+ % 2a
+ \item
+ Using the existing (English language) infrastructure to process foreign
+ queries might work better than one might expect. A lot of languages
+ share linguistic structures with English such as word positioning.
+ Moreover a lot of specialised domain words and proper names in foreign
+ languages are borrowed from English.
+
+ Of course there are also very major problems. A very big problem would
+ be not translating certain (question) words. For example the query
+ \emph{Stierf Micheal Jackson in 2009?}. When we put this query in the
+ engine it will know that something happened to \emph{MJ} in 2009 but it
+ will not know whether that something is the same as what the user
+ wanted to ask which leads to confusion.
+
+ Moreover, there are several seemingly simple structural
+ divergences (Section 25.1.2) that can cause major problems when not
+ translating such as date notation.
+
+ In conclusion, using no translation, when the language is similar to
+ English it might yield surprisingly good results. However, when the
+ difference is bigger especially the question classification will be
+ wrong and that will result into strange answers.
+
+ % 2b
+ \item
+ Translated material is hardly ever exactly the same as the original
+ materials, it either has more details that were not in the original
+ query, less details or wrong details.
+
+ More details can occur because of lexical gaps (Section 25.1.3). Some
+ language might have been developed in a region where there hardly any
+ fish, such as in the desert, and therefore the need for specialised
+ words in fishing was not there. Maybe this language only has one word
+ for fish whereas English has many. In this way extra details can be
+ inserted. Of course this also works the other way around. A popular,
+ dubious statement is often made that some Inu{\"\i}t language has over a
+ hundred words for snow. When such a specialised word is used it might
+ not be possible to correctly translate it at all to English and
+ therefore we lose detail.
+
+ % 2c
+ \item
+ The quality of the knowledge extraction depends heavily on the user's
+ language because of the aforementioned lexical gaps. However, these
+ lexical gaps might be bridged with a suitable translation system.
+\end{enumerate}
--- /dev/null
+\begin{enumerate}[label=\alph*.]
+ % 3a
+ \item
+ The \emph{Levenshtein} algorithm for edit distance is a very usefull
+ tool to detect spelling variants, however there are certain situations
+ where it will not work out of the box. One of such cases is when there
+ is a difference in script. Transliteration between scripts often
+ introduces extra letters.
+
+ For example the russian form of
+ \emph{Muhammad} becomes \emph{Mukhammed}. The \emph{kh} is a
+ construction that is not used in the English language but it sound a
+ lot like the \emph{ch} in the Scottish \emph{loch}. Such added
+ characters can introduce higher edit distances. We can possibly
+ overcome this problem by using a broader notion of characters and look
+ at phonemes for example.
+
+ \emph{Viterbi} on the other hand
+
+
+ % 3b
+ \item
+
+
+ % 3c
+ \item
+
+\end{enumerate}
repositories / itlast1617.git / commitdiff