ThyForKnowIIntroductionAbout MnemonicsWhy Mnemonics WorkMnemonic Schemes ImplementedCorporaTrainingLanguage ModelLatticeViterbiWe next use our language model to select the most likely path, which will hopefully be a well-formed natural language utterance that a person will find easy to remember. Because our lattice in this case is far from a freeform graph, we used its structure to speed up computation. For this simplest of cases, where the lattice is simply a table of word lists, we did the obvious thing. We kept the N most probable paths in a priority heap, and processed the levels of the lattice one at a time, trying each potential word with each path so far, scoring the new extended paths and remembering the best N of those. We avoided generating the lattice a priori, instead having a list of constraints along which our search advanced. Given the lexicon, our current constraint (e.g. the word must begin with ‘a’, have three syllables, and rhyme with ‘-ing’) would return to us a list of candidate words.Each path is assigned the probability using our language model. We use the usual Markov conditional independence assumption, thus given a path (w1,w2,…,wn) its probability is given byInteractivityConclusionReferencesDaniel GindikinPierre Poncecs224n/linguistics237Final ProjectAutomatic Mnemonic Device GenerationIntroductionMnemonic devices are useful memory aids that can be applied to many different aspects of daily life. There are always certain tidbits of information that people just can’t seem to remember in the form they are presented. People rely on mnemonics when they create some association between that information they wish to remember and other concepts that they already know, or find easier to remember. There are countless types of associations possible for fulfilling this purpose. People will tend to rely on the technique they feel will more easily enable them to recall the information they want at a later time.Concepts of statistical natural language processing can be used in creating a system that will automatically generate mnemonic devices. This seems like a natural application for a subject that relies on analyzing the properties of language, especially since some mnemonics are used for the express purpose of expanding one’s vocabulary. The system as implemented creates mnemonics for two distinct types of information: 1. Numerical data and2. Lexical data. Each mnemonic is represented as an utterance that is generated to present data in a way that might help the user in recall. About MnemonicsThe use of mnemonic devices dates back to the fifth century, BCE. It is recorded that Simonides of Ceos used a technique that associated specific items with locations that were well-known to the person recalling the information [1]. The effectiveness of this method of memory recall has been confirmed through a number of studies where subjects link new information to different locations [2,3]. Similar improvements in recall have been shown to materialize by using music as a mnemonic device [4].The use of mnemonics has also been shown to be of significant value in the field of education. Manalo has shown that instruction using a specific type of mnemonics known as “process mnemonics” produced improvements of mathematical ability in students classified as learning disabled [5]. Process mnemonics are used specifically for remembering rules and procedures. The mnemonics used in this system are used to allow the user to easily remember numerical and lexical information. Mnemonics for these types of information can be generated through the use of key NLP concepts. Sentencescan be created that exploit the properties of items to be remembered. A classic example of numerical information is the collection of digits found in the value of . This system can produce sentences whose word lengths are determined by the digits to be remembered.Lexical information can be coded by using the letters of an unknown word and forming a sentence whose words’ first letters match those letters. The sentence would then include one or more clue words that hint at the meaning of the unknown word. The following examples illustrate the two types of mnemonics produced by the system: : And I know, these daughters of Canaan shall see. 3 1 4 1 5 9 2 6 5 3quiescence: Quickness Us Into Everlasting StillnessWhy Mnemonics WorkMnemonics work by using the associational nature of the human memory system, as well as the differences in the ease with which various things are remembered based on their semantic categorization and relatedness. There are many various types of mnemonics depending on what is being memorized, but they all usually leverage the relative ease with which people remember spatial or natural language information, as well as the abilityto guide the recall using various constraints. For example, people are not very good at memorizing ordered sequences of unrelated objects; for instance starsare classified by their spectra into seven categories in order of decreasing temperature: O, B, A, F, G, K and M. Rather than memorizing this unrelated sequence of letters its much easier to remember the mnemonic “oh be a fine girl, kiss me”, where the first letter of each word gives the desired sequence.Mnemonic Schemes ImplementedWe implemented generation of two different types of mnemonics, both using the same algorithm. The first one was for memorizing numbers, where each individual digit is encoded in the length of words of an utterance, with zero coded for by any word longer than nine characters. For instance the speed of light, 299,792,458 m/s, can be encoded as ‘He certainly presented himself incapable of such cruel behavior’ (this particular example was generated using the Jane Austen training corpus).The second scheme was to assist with memorization of vocabulary words. We only considered words whose meanings could be captured by a single synonym, such as ‘obstinacy == tenacity’. We further cherry-picked onlythose words that contained the first letter of the synonym word, ‘obstinacy’ forexample contains ‘t’. We then formed an utterance that ended with the synonym word, where the first letters of each word formed a prefix of the wordbeing defined. For our ‘obstinacy’ example, ‘oh but such tenacity’ is such an utterance (also generated from the Jane Austen corpus). If the synonym
View Full Document
Unlocking...