For this first graph, I wanted to see how often the seven continents are mentioned in English text. I employed the case-sensitive tool to ensure that they are using these words in a continential sense. I found that Europe is mentioned far more frequently in English, American English, and British English text than any other continent, which I found very surprising.
For this second graph, I wanted to see how often love, and some of its synonyms are referred to in English fiction. The advance search tool I used was Ngram compositions to combine different Ngrams together. That way, I was able to see how often love, desire, passion, and lust were mentioned together and how often each word was mentioned. I found, as I assumed, that love was must significantly more than its counterparts in English fiction text.
I pick the book "Souls of Black Folk" by W.E.B. DuBois for the Voyant analysis. One of the tools that I found most useful was the Summary tool. This tool displays the total word and unique word count of the corpus. It also displays the vocabulary density, the average word per sentence count, and the top 5 most frequent words in the text. Another tool that was fairly insightful was the TermBerry tool. This tool presents a number of terms. When a term is hovered over, other words that co-occur with it light up as well. With the TermBerry tool you see the other words from the text that are associated with the selected word. My favorite tool to use was the Veliza tool. It was somewhat like a Siri tool, in that you can interact and communicate with Veliza. This tool also had a function that would select random lines from the uploaded text and Veliza would react to the text. It was fun to play around with it.
Words that can be positive or negative:
Words with incorrect weighting:
Both agree and correct:
Both agree but incorrect:
Disagree:
Google Translate: Incorrect Google Translate: Incorrect English --> French English --> French I love you --> Je vous aime Costs an arm and a leg --> Coûter un bras et une jambe French --> English French --> English Je vous aime --> I love you Coûter un bras et une jambe --> Costs an arm and a leg Deepl: Correct Deepl: Incorrect English --> French English --> French I love you --> Je t'aime Costs an arm and a leg --> Coûte un bras et une jambe French --> English French --> English Je t'aime --> I love you Coûter un bras et une jambe --> Costs an arm and a leg Real translation: Google Translate:Grammatically Correct (not how people usually say it) English --> French English --> Arabic Costs an arm and a leg --> Coûter les yeux de la tête I love you --> انا احبك French --> English Arabic --> English Coûter les yeux de la tête --> Costs the eyes of the head انا احبك --> I love you Bing: Correct Google Translate: Incorrect English --> Arabic English --> Arabic I love you --> أحبك Too much of something is just as bad as too little of it --> الكثير من شيء ما سيء مثل القليل جدًا منه Arabic --> English Arabic --> English أحبك --> I love you الكثير من شيء ما سيء مثل القليل جدًا منه --> Too much of something is just as bad as too little of it Bing: Incorrect Real translation: English --> Arabic English --> Arabic Too much is the same as too little --> الكثير من شيء سيء مثل القليل جدا منه Too much of something is just as bad as too little of it --> الزائد أخو الناقص Arabic --> English Arabic --> English الكثير من شيء سيء مثل القليل جدا منه --> Too much is the same as too little الزائد أخو الناقص --> Excess is the brother of shortage
Overall, these services did not do the languages justice, specifically with idioms and expressions. In theory these can be useful in practice, but if you say these to a native speaker, they'll either be confused or they'll know you aren't native speaker. It seems like Google Translate may not be the best service (least accurate), even for non-idiomatic expressions. But even then, these services sometimes work better in specific languages, so that's something to consider.
For the first experiment I did I took pictures of myself wearing no hat and then wearing various different color hats. Sadly, I think the machine may have trouble distinguishing between similar colors. I had a red and pink hat and no matter how many pictures I used or how many trainings I did, the machine has a hard time distinguishing the two. After thing, when I wasn't wearing a hat, it would often tell me I was wearing the pink or red hat with a high confidence percent (this may be because of my hair color). One thing I found very interesting, was that the percentage would flucate depending on wear I was in the frame, which I did not expect. Sometimes the machine would have a higher confidence percent of me wearing the red and/or pink hat than its percentage of me not wearing a hat (see image). Overall, I didn't find this specific experiment to be very accurate.
For the second experiment, I used the pose feature. This was way more accurate than the image recognition. It was able to distingush my poses with 100% confidence each time. This was interesting because I uploaded half the amount of pictures I uploaded in the first experiment, yet I recieved better results on the first try. I did have one pose it didn't recognize, but that was because it was extremely similar to another pose I was doing, so I had to delete that from the training. Other than that, the second experiment had no bumps.